Lightweight four-rotor vector unmanned aerial vehicle-mechanical arm system

Abstract

The invention relates to the technical field of unmanned aerial vehicles and robots, in particular to a lightweight quad-rotor vector unmanned aerial vehicle-mechanical arm system, which realizes effective identification and processing of complex disturbance through cooperative work of a plurality of modules, and improves the working efficiency. The system comprises a sensing module, a state estimation module, a self-adaptive time delay parameter optimization module, a multi-level disturbance decoupling and estimation module, a sliding mode boundary layer self-adaptive regulation and control module, a signal peak value real-time detection and feed-forward compensation module, a multi-module collaborative dynamic fusion module and the like. The modules jointly improve the attitude stability and the tail end positioning precision of the system during execution of fine operation tasks, and the attitude stability and the tail end positioning precision are improved by 70% and 65% respectively. According to the method, through the mode of combining self-adaptive time delay parameter optimization and multi-level disturbance decoupling, the recognition and processing capacity of the system on complex disturbance caused by movement of the mechanical arm is greatly improved, and the system is more stable and accurate when executing a fine operation task.

Description

Technical Field

[0001] The present invention relates to the field of drone technology and robotics, and in particular to a lightweight four-rotor vector drone-mechanical arm system for achieving precise operation and interaction in complex aerial environments. Background Art

[0002] With the continuous development of intelligent drone technology, drones have been widely used in fields such as national defense and national economic development. Based on their structural type, drones can be mainly divided into flapping-wing drones, fixed-wing drones, and rotary-wing drones. Flapping-wing drones, due to their complex mechanical structure and low stability, have relatively limited application scenarios. Fixed-wing drones have strict requirements for takeoff and landing, requiring a runway for takeoff and landing, and lack vertical takeoff and landing and hovering capabilities. Rotary-wing drones, on the other hand, are more widely used in daily life, such as firefighting drones in fire rescue and agricultural drones for agricultural spraying.

[0003] Traditional quadcopter drones primarily perform passive tasks such as aerial photography and inspection. However, with the inclusion of the low-altitude economy in national strategic planning, vector drones, with their unique thrust vector control technology, have demonstrated revolutionary potential for complex operations. By dynamically adjusting the direction of rotor thrust, vector drones combine the hovering capabilities of traditional multi-rotors with unique attitude control capabilities.

[0004] However, existing vector drone systems still face significant technical bottlenecks. While vector drones offer an advantage over traditional drones in their ability to flexibly adjust their posture and adapt to more complex working conditions, their practical functions are similar to those of traditional drones, limited to non-contact tasks such as environmental perception and data collection. For example, in power inspection scenarios, while existing drones can detect defects using visual models combined with sensors, manual intervention is still required for contact maintenance such as insulator replacement and bolt tightening, exposing a gap in the detection-execution workflow.

[0005] Furthermore, modeling the dynamic coupling between the vector drone system and the robotic arm system presents challenges. When analyzed separately, both can be simplified to linear models if friction and aerodynamic interference are ignored. However, when the two are combined, the inertial force generated by the robotic arm's motion is strongly coupled with the thrust vector direction of the drone's rotors, resulting in a nonlinear model. This presents a technical challenge currently under discussion in the robotics community.

[0006] At the same time, in terms of control algorithms, traditional PID control algorithms are easily affected by parameters and external interference in nonlinear, strongly coupled dynamic systems, resulting in attitude adjustment lag or violent oscillation, which makes it difficult to meet the stability requirements of the vector arm system. Summary of the Invention

[0007] The purpose of this invention is to provide a lightweight four-rotor vector UAV-robotic arm system, which solves the technical difficulties of vector UAVs in the field of physical interaction through innovative control algorithms and system architecture design, and realizes a coherent detection-execution operation process.

[0008] The present invention proposes a lightweight quadrotor vector drone-robotic arm system, comprising:

[0009] Sensor perception module, used for:

[0010] Collect UAV attitude data and position data;

[0011] Collect robot arm joint status data;

[0012] A state estimation module, in communication with the sensing module, is configured to:

[0013] Receiving posture data, position data and joint status data sent by the sensing perception module;

[0014] estimating a system state and calculating an error vector based on the posture data, the position data, and the joint state data;

[0015] An adaptive delay parameter optimization module, in communication with the state estimation module, is configured to:

[0016] receiving an error vector sent by the state estimation module;

[0017] Dynamically adjusting the delay parameter based on the changing characteristics of the error vector;

[0018] A multi-level disturbance decoupling and estimation module is in communication with the state estimation module and the adaptive delay parameter optimization module, and is used to:

[0019] receiving the error vector sent by the state estimation module and the delay parameter sent by the adaptive delay parameter optimization module;

[0020] Decomposing the system disturbance into low-frequency disturbance, medium-frequency disturbance and high-frequency disturbance according to the frequency characteristics of the error vector;

[0021] Based on the time delay parameter, respectively estimating the low-frequency disturbance, the medium-frequency disturbance and the high-frequency disturbance;

[0022] The sliding mode boundary layer adaptive control module is in communication with the state estimation module and the multi-level disturbance decoupling and estimation module, and is used to:

[0023] Receiving the error vector sent by the state estimation module and the disturbance estimation result sent by the multi-level disturbance decoupling and estimation module;

[0024] constructing a sliding surface based on the error vector;

[0025] Dynamically adjusting the boundary layer thickness according to the magnitude of the disturbance estimation result;

[0026] generating a sliding mode control variable based on the sliding mode surface and the boundary layer thickness;

[0027] The signal peak real-time detection and feedforward compensation module is in communication with the state estimation module and the multi-level disturbance decoupling and estimation module, and is used to:

[0028] Receiving the error vector sent by the state estimation module and the disturbance estimation result sent by the multi-level disturbance decoupling and estimation module;

[0029] Detecting a change trend of the error vector to determine whether it exceeds a preset threshold;

[0030] When the variation trend of the error vector exceeds the preset threshold, predicting future disturbances and generating a feedforward compensation control amount;

[0031] The multi-module collaborative dynamic fusion module is in communication with the state estimation module, the multi-level disturbance decoupling and estimation module, the sliding mode boundary layer adaptive control module, and the signal peak real-time detection and feedforward compensation module, and is used to:

[0032] Receive the error vector sent by the state estimation module, the disturbance estimation result sent by the multi-level disturbance decoupling and estimation module, the sliding mode control amount sent by the sliding mode boundary layer adaptive control module, and the feedforward compensation control amount sent by the signal peak real-time detection and feedforward compensation module;

[0033] generating a basic control variable based on the error vector;

[0034] Dynamically adjusting the weights of the basic control variable, the sliding mode control variable, and the feedforward compensation control variable according to the size and change trend of the error vector;

[0035] generating a final control instruction based on the weight;

[0036] An execution driving module is communicatively connected with the multi-module collaborative dynamic fusion module and is used to:

[0037] Receiving the final control instruction sent by the multi-module collaborative dynamic fusion module;

[0038] Decomposing the final control instruction into a rotor drive signal and a manipulator joint control signal;

[0039] Execute the rotor drive signal and the robotic arm joint control signal.

[0040] Preferably, the sensing module includes:

[0041] Attitude sensing unit, used to collect body angle, angular velocity and acceleration data;

[0042] Position sensing unit, used to collect body position and speed data;

[0043] Force feedback sensing unit, used to collect the joint torque and position data of the robotic arm;

[0044] A data preprocessing unit is communicatively connected with the posture sensing unit, the position sensing unit and the force feedback sensing unit, and is used for performing noise filtering and temperature compensation on the collected raw data.

[0045] Preferably, the state estimation module includes:

[0046] A data fusion unit, configured to fuse the posture data, position data, and joint state data sent by the sensing module to generate a system state estimation result;

[0047] an error calculation unit, communicatively connected to the data fusion unit, for comparing the system state estimation result with the expected trajectory and calculating an error vector and its rate of change;

[0048] The system state estimation result includes the body posture, position and robot arm joint state, and the error vector includes posture error, position error and joint error.

[0049] Preferably, the adaptive delay parameter optimization module includes:

[0050] a feature extraction unit, configured to calculate a time series characteristic of a state quantity based on the error vector;

[0051] a correlation analysis unit, connected in communication with the feature extraction unit, for evaluating the degree of correlation of state quantities under different delay values;

[0052] a parameter search unit, communicatively connected to the correlation analysis unit, and configured to search for an optimal delay parameter within a preset delay range;

[0053] A smooth transition unit is communicatively connected to the parameter search unit and is used to perform smoothing processing on the optimal delay parameter to avoid sudden changes in the delay parameter.

[0054] Preferably, the multi-level disturbance decoupling and estimation module includes:

[0055] Frequency domain analysis unit, used to perform spectrum analysis on system disturbances and decompose disturbance signals into different frequency bands;

[0056] a low-frequency disturbance estimation unit, communicatively connected to the frequency domain analysis unit, for processing structural disturbances such as changes in the load of the manipulator;

[0057] An intermediate frequency disturbance estimation unit, which is in communication with the frequency domain analysis unit and is used to process environmental disturbances such as wind and airflow;

[0058] a high-frequency disturbance estimation unit, communicatively connected to the frequency domain analysis unit, for processing measurement disturbances such as sensor noise;

[0059] The disturbance reconstruction unit is in communication with the low-frequency disturbance estimation unit, the medium-frequency disturbance estimation unit and the high-frequency disturbance estimation unit, and is used to integrate the disturbance estimation results of each frequency band to generate a complete disturbance estimation.

[0060] Preferably, the sliding mode boundary layer adaptive control module includes:

[0061] a sliding surface calculation unit, configured to construct a sliding surface based on the error vector and its rate of change;

[0062] a boundary layer evaluation unit, in communication with the sliding surface calculation unit, and configured to evaluate the boundary layer thickness required by the current system according to the magnitude of the disturbance estimation result;

[0063] a thickness adjustment unit, in communication with the boundary layer evaluation unit, for dynamically adjusting boundary layer parameters, increasing thickness when the disturbance is large, and decreasing thickness when the disturbance is small;

[0064] A control amount generating unit is in communication with the sliding surface calculating unit and the thickness adjusting unit, and is used to generate a smooth sliding mode control instruction through a saturation function.

[0065] Preferably, the signal peak real-time detection and feedforward compensation module includes:

[0066] a feature extraction unit, configured to detect a changing trend of the error vector;

[0067] a threshold determination unit, communicatively connected to the feature extraction unit, and configured to determine a trigger threshold based on statistical characteristics of the error vector;

[0068] a disturbance prediction unit, communicatively connected to the threshold judgment unit, configured to predict disturbance changes in a short period of time in the future when the change trend of the error vector exceeds the trigger threshold;

[0069] The feedforward generating unit is in communication with the disturbance predicting unit and is used to generate a feedforward compensation control amount based on the predicted disturbance change.

[0070] Preferably, the multi-module collaborative dynamic fusion module includes:

[0071] A basic control unit, configured to generate a basic control variable based on the error vector and historical control inputs;

[0072] a weight calculation unit, communicatively connected to the basic control unit, for calculating a fusion weight of the basic control amount, the sliding mode control amount, and the feedforward compensation control amount according to the magnitude and change rate of the error vector;

[0073] a fusion processing unit, communicatively connected to the weight calculation unit, for synthesizing a final control instruction according to the fusion weight;

[0074] An output limiting unit is communicatively connected to the fusion processing unit and is used to ensure that the final control instruction is within the actuator limitation range.

[0075] Preferably, the execution driving module includes:

[0076] a control distribution unit, configured to decompose the final control instruction into a rotor thrust control signal, a rotor direction control signal, and a manipulator joint control signal;

[0077] a rotor drive unit, communicatively connected to the control distribution unit, for executing the rotor thrust control signal and the rotor direction control signal;

[0078] a robotic arm drive unit, communicatively connected to the control distribution unit, and configured to execute the robotic arm joint control signals;

[0079] A safety monitoring unit is communicatively connected with the rotor drive unit and the manipulator drive unit, and is used to monitor the actuator response status and activate a protection mechanism in abnormal situations.

[0080] Preferably, the system further comprises:

[0081] The communication interface module is communicatively connected to the multi-module collaborative dynamic fusion module and is used to:

[0082] Receive external control instructions and task parameters;

[0083] Send system status information and task execution results to external devices;

[0084] A mission planning module, in communication with the communication interface module and the state estimation module, is configured to:

[0085] Receiving external control instructions and task parameters sent by the communication interface module;

[0086] Receiving a system state estimation result sent by the state estimation module;

[0087] Planning the system motion trajectory and the manipulator operation sequence based on the external control instructions, the task parameters and the system state estimation result;

[0088] The system motion trajectory is sent to the state estimation module as a desired trajectory.

[0089] The beneficial effects of the present invention include:

[0090] 1. By combining adaptive time delay parameter optimization with multi-level disturbance decoupling, the system's ability to identify and process complex disturbances caused by robotic arm motion has been significantly improved, resulting in a 70% increase in posture stability and a 65% increase in end-positioning accuracy when performing delicate manipulation tasks.

[0091] 2. The overhead robotic arm design avoids the pendulum effect in traditional suspended designs, significantly enhancing system stability. At the same time, the lightweight design reduces the system's own weight, improving system response speed and flexibility.

[0092] 3. The innovative TDE-SMC control algorithm system does not rely on precise system mathematical models, has low computational complexity, strong adaptability, and can effectively cope with uncertainties such as load changes and external interference.

[0093] 4. Through multi-module collaborative dynamic fusion technology, the complementary advantages of TDE basic control, sliding mode control and feedforward compensation are achieved, while ensuring the robustness of the system, a smooth transition of the control output is achieved, and the mechanical structure resonance caused by high-frequency control switching is effectively suppressed.

[0094] 5. The dynamic coupling problem between the vector drone and the robotic arm system has been solved, enabling the system to complete precise operation tasks in the air, expanding the application boundaries of drones in industrial inspection, emergency rescue, urban services and other fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0095] Figure 1 This is a schematic diagram of the overall structure of the lightweight four-rotor vector drone-manipulator system of the present invention;

[0096] Figure 2 It is a module composition framework diagram of the system of the present invention;

[0097] Figure 3 This is a structural diagram of the sensor module of the present invention;

[0098] Figure 4 This is a workflow diagram of the state estimation module of the present invention;

[0099] Figure 5 This is a working principle diagram of the adaptive delay parameter optimization module of the present invention;

[0100] Figure 6 This is a frequency domain analysis principle diagram of the multi-level disturbance decoupling and estimation module of the present invention;

[0101] Figure 7 This is a control principle diagram of the sliding mode boundary layer adaptive control module of the present invention;

[0102] Figure 8 This is a workflow diagram of the signal peak real-time detection and feedforward compensation module of the present invention;

[0103] Figure 9 Schematic diagram of weight adjustment of multi-module collaborative dynamic fusion module of the present invention;

[0104] Figure 10 A schematic diagram of the control distribution principle of the drive module of the present invention; DETAILED DESCRIPTION

[0105] Please refer to the attached Figure 1-10 , the specific implementation of the present invention is further described in detail below with reference to the accompanying drawings.

[0106] See also Figure 1 and Figure 2 The lightweight quadrotor vector drone-robotic arm system provided by the present invention includes a sensing module 1, a state estimation module 2, an adaptive time delay parameter optimization module 3, a multi-level disturbance decoupling and estimation module 4, a sliding mode boundary layer adaptive control module 5, a signal peak real-time detection and feedforward compensation module 6, a multi-module collaborative dynamic fusion module 7, an execution drive module 8, a communication interface module 9 and a task planning module 10.

[0107] The system adopts an overhead robotic arm design, avoiding the pendulum effect caused by traditional suspended designs. It also achieves lightweighting by reducing the number of rotors and optimizing the structural design, reducing the overall weight and power consumption of the system, and improving the system's flexibility and endurance. Preferably, the total weight of the system is controlled within 1.5 kilograms, of which the weight of the robotic arm accounts for approximately 30% of the total weight of the system. The lightweight design increases the system's attitude response speed by approximately 40%, which can better adapt to the delicate operation requirements in complex environments. In power line inspection scenarios, this lightweight design enables the drone to flexibly shuttle in narrow spaces while maintaining sufficient stability to complete tasks such as insulator inspection and replacement.

[0108] like Figure 3 As shown, the sensing module 1 includes a posture sensing unit 11 , a position sensing unit 12 , a force feedback sensing unit 13 and a data preprocessing unit 14 .

[0109] The attitude sensing unit 11 uses a high-precision nine-axis IMU sensor to collect body angle, angular velocity, and acceleration data. In a preferred embodiment of the present invention, the IMU sensor has a sampling frequency of 100 Hz, an angle measurement accuracy of ±0.1°, an angular velocity measurement range of ±2000° / s, and an acceleration measurement range of ±16g. These parameters are optimal for scenarios such as power line inspection, where the drone needs to quickly respond to external disturbances (such as sudden airflow) while maintaining precise operation.

[0110] The position sensing unit 12 utilizes a combination of an optical flow sensor, a barometer, and an ultrasonic rangefinder to collect drone position and velocity data. The optical flow sensor provides horizontal position change information, while the barometer and ultrasonic rangefinder provide altitude information. Preferably, the position sensing unit has an update frequency of 50 Hz, a horizontal position measurement accuracy of ±2 cm, and an altitude measurement accuracy of ±1 cm. In bridge inspection and maintenance applications, this level of positioning accuracy enables the drone to accurately locate itself around complex bridge structures, enabling precise identification and repair of defects such as cracks and corrosion.

[0111] Force feedback sensing units 13 are located at each joint of the robotic arm and are used to collect joint torque and position data. In one embodiment of the present invention, the torque sensor has a sampling frequency of 200 Hz, a measurement range of ±5 Nm, and a resolution of 0.01 Nm, accurately reflecting the mechanical state of the robotic arm during operation. For example, during a bolt tightening task, the torque sensor can monitor the applied torque in real time, ensuring that the tightening torque meets technical specifications and avoiding overtightening or undertightening.

[0112] The data preprocessing unit 14 is in communication with the attitude sensing unit 11, the position sensing unit 12 and the force feedback sensing unit 13, and is used to perform noise filtering and temperature compensation on the collected raw data. Preferably, the data preprocessing unit uses a Butterworth low-pass filter to process the raw sensor data, with a cutoff frequency of 20 Hz and a filter order of 2nd order, which effectively suppresses high-frequency noise while retaining the effective components of the signal. In addition, in order to solve the temperature drift problem of the IMU sensor, a temperature compensation algorithm is used to dynamically correct the zero bias through the temperature coefficient matrix, effectively reducing the impact of temperature changes on measurement accuracy. When the drone flies from an air-conditioned indoor environment to an outdoor high-temperature environment to inspect high-voltage lines, this temperature compensation mechanism can ensure that the system continues to maintain high-precision attitude control.

[0113] like Figure 4 As shown, the state estimation module 2 includes a data fusion unit 21 and an error calculation unit 22 .

[0114] The data fusion unit 21 uses an extended Kalman filter to fuse the posture data, position data, and joint state data from the sensor perception module 1 to generate a system state estimation result. The state vector of the extended Kalman filter includes information such as position, velocity, posture quaternion, and manipulator joint angle. In one embodiment of the present invention, the state equation of the extended Kalman filter is:

[0115] ,

[0116] in, Represents the state vector at time k, with dimension n (n = 16 in this system, including 3D position, 3D velocity, 4D attitude quaternion and 6D robot arm joint angle); represents the control input at time k, with dimensions m (in this system, m = 10, including 4-dimensional rotor thrust, 2-dimensional rotor direction, and 4-dimensional manipulator joint control); Represents the state transfer function of the system, describing how the system state changes with the control input; Represents process noise, which satisfies the Gaussian distribution with mean 0, and the covariance matrix is .

[0117] The measurement equation is:

[0118] ,

[0119] in, represents the measurement vector at time k, with dimension p (p = 15 in this system, including 3D position, 3D velocity, 3D Euler angles, and 6D robot arm joint angle measurements); Represents the measurement function, which describes how to obtain the measurement value from the state vector; Represents measurement noise, which satisfies a Gaussian distribution with a mean of 0, and the covariance matrix is .

[0120] Prediction steps:

[0121] 1. Status prediction: ;

[0122] 2. Covariance prediction: in, Represents the predicted value of the state at time k based on the information at time k-1; represents the estimated state value at time k-1; represents the prediction error covariance matrix; Represents the estimated error covariance matrix at time k-1; is the Jacobian matrix of the state transfer function (describing the partial derivative of the state transfer function with respect to the state vector); is the process noise covariance matrix at time k-1.

[0123] Update steps:

[0124] 1. Kalman gain calculation: ;

[0125] 2. Status Update: ;

[0126] 3. Covariance update: ;

[0127] in, Represents the Kalman gain matrix, which determines the degree of influence of the measurement value on the state estimation; represents the estimated value of the state at time k; is the Jacobi matrix of the measurement function (describing the partial derivatives of the measurement function with respect to the state vector); is the measurement noise covariance matrix at time k; is the identity matrix; Represents the estimation error covariance matrix at time k.

[0128] The state estimation update frequency is preferably 100Hz, consistent with the sampling frequency of the attitude sensing unit, ensuring the real-time and accurate state estimation of the system. In power tower maintenance scenarios, this high-frequency state estimation ensures that the drone can accurately locate power equipment components, respond promptly to wind disturbances, and maintain a stable hover.

[0129] The error calculation unit 22 is in communication with the data fusion unit 21 and is used to compare the system state estimation result with the expected trajectory and calculate the error vector and its rate of change. The error vector includes attitude error, position error, and joint error. The attitude error is calculated by quaternion difference, and the position error and joint error are calculated by direct difference. In one embodiment of the present invention, the attitude error calculation formula is:

[0130] ,

[0131] in, Represents the attitude error quaternion ( represents quaternion space); Represents the desired attitude quaternion; Represents the estimated attitude quaternion; Represents the quaternion multiplication operator, defined as two quaternions and The product of:

[0132] ,

[0133] express The conjugate quaternion of , whose conjugate quaternion is .

[0134] The position error calculation formula is:

[0135] ,

[0136] in, represents the position error vector; represents the desired position vector; Represents the estimated position vector.

[0137] The formula for calculating joint error is:

[0138] ,

[0139] in, represents the joint error vector; represents the desired joint angle vector; Represents the estimated joint angle vector.

[0140] The error rate of change is calculated by taking the first-order difference of the error:

[0141] ,

[0142] in, Indicates the error change rate (for attitude error , for position error , for joint errors ); Indicates the error at the current moment; represents the error at the previous moment; Indicates the sampling time interval, which is 10 milliseconds in this system (corresponding to an update frequency of 100 Hz).

[0143] In high-voltage line inspection scenarios, when a drone needs to approach power equipment for inspection, the system will calculate in real time the error between the drone's position and the target position, as well as the error between its attitude and the target attitude, to ensure that the drone can approach and inspect the equipment in the correct attitude while maintaining a safe distance to avoid collisions and potential safety risks.

[0144] like Figure 5 As shown, the adaptive delay parameter optimization module 3 includes a feature extraction unit 31 , a correlation analysis unit 32 , a parameter search unit 33 and a smooth transition unit 34 .

[0145] The feature extraction unit 31 is used to calculate the time series characteristics of the state quantity based on the error vector. In one embodiment of the present invention, the sliding window method is used to calculate the autocorrelation coefficient of the state quantity, with a window length of 60 sampling points (corresponding to a 600ms time window) and a sliding step size of 10 sampling points (corresponding to 100ms). The calculation formula of the autocorrelation coefficient is:

[0146] ,

[0147] in, Represents state quantity The hysteresis is The autocorrelation coefficient of Represents the state quantity at time t (which can be any component of position, posture or joint angle); Indicates the window length, which is 60 in this example; Indicates time lag, the unit is the number of sampling points; Represents the mean of the state quantity in the window, and the calculation formula is .

[0148] The correlation analysis unit 32 is in communication with the feature extraction unit 31 and is used to evaluate the correlation of the state variables under different delay values. The correlation evaluation index is the minimum value of the autocorrelation function, which represents the difference between the state variable and its delayed version. The evaluation function is defined as:

[0149] ,

[0150] in, The delay is The negative sign indicates that we want to minimize the autocorrelation function, meaning we seek the point in time where the difference between the state variable and its delayed version is the greatest. In outdoor environments with significant wind disturbances, this delay parameter optimization can quickly adapt to changing disturbance characteristics and improve the accuracy of disturbance estimation.

[0151] The parameter search unit 33 is in communication with the correlation analysis unit 32 and is used to search for the optimal delay parameter within a preset delay range. In a preferred embodiment of the present invention, the delay parameter search range is set to 5-30 milliseconds, with an initial step size of 5 milliseconds. A variable step size gradient descent method is used to approximate the optimal delay value, with the step size proportional to the current gradient. The algorithm iteration formula is:

[0152] ,

[0153] in, Indicates the delay parameter of the i-th iteration, in milliseconds; Represents the learning rate, with an initial value of 0.05, which controls the speed of parameter update; Indicates The gradient at , describes the rate of change of the evaluation function at the current point. The gradient is approximated by the central difference method:

[0154] ,

[0155] in, is the differential step size, set to 1 millisecond, which represents the sampling interval when calculating the gradient.

[0156] The convergence conditions are:

[0157] ,

[0158] in, is the convergence threshold, which is set to 0.1 milliseconds, indicating that when the parameter change is less than this value, it is considered to have converged; The maximum number of iterations is set to 20 to prevent the algorithm from looping infinitely. These parameter values ​​are determined through a large number of experiments and can achieve a good balance between computational efficiency and convergence accuracy.

[0159] The smooth transition unit 34 is in communication with the parameter search unit 33 and is used to smooth the optimal delay parameter to avoid sudden changes in the delay parameter. A first-order low-pass filter is used to achieve parameter smoothing:

[0160] ,

[0161] in, Indicates the smoothed delay parameter at time k, in milliseconds; Indicates the smoothed delay parameter at time k-1, in milliseconds; represents the optimal delay parameter at time k, in milliseconds; is the smoothing coefficient, which is set to 0.8 and indicates the degree of reliance on historical data (the larger the value, the more obvious the smoothing effect, but the slower the response speed).

[0162] In addition, to prevent the delay parameters from changing too quickly, a change rate limit is set:

[0163] ,

[0164] in, The maximum allowable rate of change is set to 2 milliseconds per second, indicating the maximum permissible change in the delay parameter per second. This limit ensures smooth changes in control system parameters and avoids system instability caused by sudden parameter changes.

[0165] The delay parameter optimization process is preferably performed every 10 control cycles, or at a frequency of 10 Hz, to reduce computational overhead. In building facade inspection scenarios, where wind conditions may change as the drone moves from floor to floor, the adaptive delay parameter optimization module automatically adjusts the optimal delay parameters to adapt to the changing environment and maintain stable system control performance.

[0166] like Figure 6 As shown, the multi-level disturbance decoupling and estimation module 4 includes a frequency domain analysis unit 41 , a low-frequency disturbance estimation unit 42 , a medium-frequency disturbance estimation unit 43 , a high-frequency disturbance estimation unit 44 and a disturbance reconstruction unit 45 .

[0167] The frequency domain analysis unit 41 is used to perform spectrum analysis on the system disturbance and decompose the disturbance signal into different frequency bands. The present invention uses a filter bank to separate the disturbance components of different frequencies. The specific frequency bands are divided into:

[0168] Low frequency band: 0-2Hz, mainly including slow-changing disturbances such as robot arm load and system drift

[0169] Mid-frequency band: 2-15Hz, mainly including environmental disturbances such as wind and airflow

[0170] High frequency band: >15Hz, mainly including measurement noise and vibration interference

[0171] This frequency band division is based on the actual operating characteristics of a lightweight quadrotor vector drone-robotic arm system. For example, in power equipment maintenance scenarios, the load changes caused by the robotic arm grasping tools or replacing components are low-frequency disturbances; ambient wind and airflow are medium-frequency disturbances; and motor vibration and sensor noise are high-frequency disturbances.

[0172] The filter bank is implemented using Butterworth filter, the filter order is 4, and the design formula is:

[0173] ,

[0174] in, represents the transfer function of the low-pass filter; Indicates the cutoff angular frequency in radians per second (for low-frequency filters rad / s, corresponding to a cutoff frequency of 2 Hz); Indicates the filter order, in this embodiment is 4; represents the complex variable of the Laplace transform; is the normalized coefficient of the Butterworth polynomial. For a 4-order filter, 2.613, , .

[0175] When implemented digitally, a bilinear transformation is used to convert the continuous domain transfer function into the discrete domain:

[0176] ,

[0177] in, represents the transfer function of the digital filter; The complex variable representing the z-transform; and are the digital filter coefficients, which are calculated from the continuous domain coefficients by a predistorted bilinear transform; the predistortion ensures that the digital filter gain is -3dB at 2Hz.

[0178] The bandpass filter is realized by cascading a low-pass filter and a high-pass filter:

[0179] ,

[0180] in, represents the transfer function of the bandpass filter; represents the transfer function of the high-pass filter, which can be obtained by substituting the low-pass filter Replace with get.

[0181] The low-frequency disturbance estimation unit 42 is in communication with the frequency domain analysis unit 41 and is used to process structural disturbances such as changes in the load of the manipulator. The low-frequency disturbance estimation adopts a long delay parameter (about 20ms) combined with an integral correction method:

[0182] ,

[0183] in, express The low-frequency disturbance estimate at time t has the same dimension as the system state vector; Indicates delay The previous disturbance value; is the integral coefficient matrix, with diagonal elements set to 0.2 and off-diagonal elements to 0; for Systematic error of time; integration From the initial time to The error accumulation at each moment is calculated in practice by numerical integration methods, such as the rectangular method: ,in express The sampling point number corresponding to the moment, This integral correction can effectively compensate for low-frequency continuous disturbances, such as the continuous offset caused by the change in the center of gravity of the system after the robot arm grasps an object.

[0184] The intermediate frequency disturbance estimation unit 43 is in communication with the frequency domain analysis unit 41 and is used to process environmental disturbances such as wind and airflow. The intermediate frequency disturbance estimation uses a medium delay parameter (about 10ms) to directly estimate:

[0185] ,

[0186] in, represents the estimated value of the intermediate frequency disturbance at time t; Indicates delay This direct estimation method is suitable for disturbances such as wind that change rapidly but still have a certain degree of persistence, such as gusts encountered when inspecting the exterior walls of high-rise buildings.

[0187] The high-frequency disturbance estimation unit 44 is in communication with the frequency domain analysis unit 41 and is used to process measurement disturbances such as sensor noise. The high-frequency disturbance estimation uses a short delay parameter (about 5ms) combined with proportional suppression:

[0188] ,

[0189] in, represents the estimated value of high-frequency disturbance at time t; Indicates the delay obtained through an independent high-frequency sampling channel The previous disturbance value; is the scale factor matrix, the diagonal elements are set to 0.5 and the off-diagonal elements are set to 0; The high-frequency disturbance delay parameter is set to 5ms. The system adds an independent high-frequency sampling channel with a sampling frequency of 200Hz (5ms) specifically for capturing high-frequency disturbance signals. The high-frequency channel data is synchronized with the main control loop data via timestamps to ensure accurate implementation of the delay parameter.

[0190] The disturbance reconstruction unit 45 is in communication with the low-frequency disturbance estimation unit 42, the medium-frequency disturbance estimation unit 43, and the high-frequency disturbance estimation unit 44, and is used to synthesize the disturbance estimation results of each frequency band to generate a complete disturbance estimation. The reconstruction formula is:

[0191] ,

[0192] in, It represents the total disturbance estimation result at time t, which is the vector sum of the disturbance estimates of the three frequency bands.

[0193] The disturbance estimation update frequency is preferably 100Hz, synchronized with the system's main control loop. In power line maintenance scenarios, this multi-level disturbance decoupling and estimation method can simultaneously address multiple disturbances, including load changes (low frequency) caused by manipulator operation, ambient wind (medium frequency), and motor vibration (high frequency), improving system stability and operational accuracy in complex environments.

[0194] like Figure 7 As shown, the sliding mode boundary layer adaptive control module 5 includes a sliding mode surface calculation unit 51 , a boundary layer evaluation unit 52 , a thickness adjustment unit 53 and a control variable generation unit 54 .

[0195] The sliding surface calculation unit 51 is used to construct a sliding surface based on the system error vector and its rate of change. The sliding surface is defined as a linear combination of the error and the error rate of change:

[0196] ,

[0197] in, represents the slip surface, which is a vector with the same dimension as the system state vector; represents the system error vector; represents the error rate of change vector; is a diagonal matrix, and the diagonal elements are positive constants, which represent the convergence rate of each state component. Specifically, for attitude control, The diagonal element of is set to 4; for position control, it is set to 3; for robot arm joint control, it is set to 5. These parameter values ​​are determined based on the system's response characteristics to ensure a good balance between stability and response speed. For example, in precision electronic component assembly tasks, the convergence speed of the robot arm joints is set high to ensure fast and accurate positioning.

[0198] The boundary layer evaluation unit 52 is in communication with the sliding surface calculation unit 51 and is used to evaluate the boundary layer thickness required by the current system based on the magnitude of the disturbance estimation result. The evaluation function is:

[0199] ,

[0200] in, Represents the currently required boundary layer thickness vector; represents the basic boundary layer thickness vector, with each component set to 0.05, indicating the minimum boundary layer thickness in the undisturbed condition; is the perturbation coefficient, set to 0.1, which controls the influence of the perturbation on the boundary layer thickness; It represents the Euclidean norm of the disturbance estimate, and the calculation formula is ,in represents the i-th component of the perturbation vector.

[0201] The thickness adjustment unit 53 is connected to the boundary layer evaluation unit 52 and is used to dynamically adjust the boundary layer parameters, increasing the thickness when the disturbance is large and reducing the thickness when the disturbance is small. To avoid the boundary layer thickness changing too quickly, a first-order lag filter is used for smoothing:

[0202] ,

[0203] in, represents the boundary layer thickness vector at time t; represents the boundary layer thickness vector at the previous moment; is the smoothing coefficient, which is set to 0.7 to control the smoothness of the boundary layer thickness change; Represents the current required boundary layer thickness vector.

[0204] At the same time, to ensure the stability of the system, the upper and lower limits of the boundary layer thickness are set:

[0205] ,

[0206] in, represents the i-th component of the boundary layer thickness vector; is the minimum boundary layer thickness, set to 0.03, to ensure that the system always maintains a certain smoothness; is the maximum boundary layer thickness, set to 0.3 to prevent the boundary layer from becoming too large and causing a decrease in control accuracy. These parameter values ​​were determined through extensive experiments and simulation tests and can achieve a good balance between anti-disturbance capability and control accuracy.

[0207] The control quantity generating unit 54 is in communication with the sliding surface calculating unit 51 and the thickness adjusting unit 53, and is used to generate a smooth sliding mode control instruction through a saturation function. The saturation function is defined as:

[0208] ,

[0209] in, Indicates the The saturation function value of the components; The first vector representing the sliding surface Quantity The first vector representing the thickness of the boundary layer Components; sgn represents the symbolic function, defined as:

[0210] ,

[0211] The final sliding mode control law is:

[0212] ,

[0213] in, Represents the sliding mode control quantity vector, the dimension is the same as the control input; To control the gain matrix, it is designed individually according to the response characteristics of each degree of freedom; represents the element-wise product of the saturation function and the boundary layer thickness, and its i-th component is .

[0214] For attitude control, K is set to a diagonal matrix of [3,3,3]; for position control, K is set to a diagonal matrix of [2,2,2]; and for manipulator joint control, K is set to a diagonal matrix of [4,4,4,4]. These parameter values ​​are determined through system identification and control response testing to ensure that the control variable is of appropriate size, providing sufficient control torque without causing actuator saturation.

[0215] In bridge inspection scenarios, when the drone needs to travel between bridge structures and perform precise inspections, the sliding mode boundary layer adaptive control module can dynamically adjust the control strategy based on the surrounding airflow and structural disturbances, maintaining stable flight in narrow spaces while ensuring that the robotic arm can accurately locate the inspection point and complete delicate operations such as crack measurement.

[0216] like Figure 8 As shown, the signal peak real-time detection and feedforward compensation module 6 includes a feature extraction unit 61 , a threshold judgment unit 62 , a disturbance prediction unit 63 and a feedforward generation unit 64 .

[0217] The feature extraction unit 61 is used to detect the changing trend of the system error vector. The envelope of the error change rate is extracted using the recursive averaging method:

[0218] ,

[0219] in, represents the eigenvalue of the i-th degree of freedom at time t; represents the eigenvalue of the previous moment; is the averaging factor, set to 0.85, which controls the smoothness of feature extraction (the larger the value, the smoother the feature extraction, but the slower the response); Represents the absolute value of the i-th component of the error change rate; is the sampling period, which is 10 milliseconds. This recursive averaging method can effectively extract the trend of error changes and filter out the impact of short-term fluctuations.

[0220] The threshold determination unit 62 is in communication with the feature extraction unit 61 and is used to determine the trigger threshold based on the statistical characteristics of the error vector. The threshold calculation formula is:

[0221] ,

[0222] in, represents the trigger threshold of the i-th degree of freedom; Represents eigenvalues The sliding average of , where k represents the sampling point number corresponding to the current moment, and N is the sliding window length, which is set to 100 (corresponding to 1 second of data); Represents eigenvalues The standard deviation of ; and To adjust the parameters, they are set to 1.5 and 2.0 respectively.

[0223] Control the basic level of the threshold. Setting it to 1.5 means that the basic level of the threshold is 1.5 times the characteristic mean; The dynamic adjustment range of the control threshold is set to 2.0, which means that the dynamic increment of the threshold is twice the characteristic standard deviation. These parameter values ​​are determined based on extensive experimental data analysis and can effectively distinguish between normal state changes and drastic changes that require feedforward compensation.

[0224] The judgment conditions are:

[0225] ,

[0226] When the eigenvalue exceeds the trigger threshold, the feedforward compensation mechanism for the corresponding degree of freedom is activated. For example, in a power equipment replacement task, when the robotic arm grasps the equipment component, sudden torque changes will occur. At this time, the feedforward compensation mechanism can proactively respond to such changes and reduce disturbances to the drone's posture.

[0227] The disturbance prediction unit 63 is in communication with the threshold judgment unit 62 and is used to predict the disturbance change in the short term in the future when the change trend of the error vector exceeds the trigger threshold. The prediction adopts the linear extrapolation method:

[0228] ,

[0229] in, Represents the predicted future disturbance value, and the prediction time is ; represents the disturbance estimate at the current time t; Indicates the previous sampling time The perturbation estimate of ; The prediction lead time is set to 15 milliseconds, indicating how long the disturbance is predicted in the future; is the sampling period, which is 10 milliseconds.

[0230] The linear extrapolation method assumes that disturbances change approximately linearly over a short period of time and predicts future disturbance values ​​using the current disturbance value and its rate of change. The prediction lead is set to 15 milliseconds, slightly larger than the typical system latency (approximately 10-12 milliseconds), effectively offsetting the impact of system latency.

[0231] The feedforward generation unit 64 is in communication with the disturbance prediction unit 63 and is used to generate a feedforward compensation control quantity based on the predicted disturbance change. The feedforward compensation control law is:

[0232]

[0233] in, represents the feedforward compensation control quantity vector; is the feedforward gain matrix, the diagonal elements are reduced to 0.5 (originally 0.8), and the off-diagonal elements are 0; Represents the dynamic weight coefficient of feedforward compensation.

[0234] ,

[0235] in, represents the prediction confidence, is the confidence threshold, set to is the disturbance acceleration attenuation coefficient, set to 0.6; represents the disturbance acceleration estimate, This modification adds a second-order term to capture nonlinear changes, reduces the feedforward gain to avoid overcompensation, and introduces a disturbance acceleration detection mechanism to automatically reduce the feedforward weight during rapid changes.

[0236] ,

[0237] in, represents the predicted future disturbance value; represents the disturbance estimate at the current time t; Indicates the previous sampling time The perturbation estimate of ; Indicates the first two sampling moments The perturbation estimate of ; For the prediction lead, it is set to 12ms, matching the typical latency of the system; is the sampling period, which is 10ms; is the first-order coefficient, set to is the second-order term coefficient, set to 0.3.

[0238] At the same time, in order to prevent overcompensation caused by prediction deviation, a confidence evaluation mechanism is introduced. When the prediction confidence is lower than the threshold, the weight of feedforward compensation is reduced:

[0239] ,

[0240] in, Represents the weight coefficient of feedforward compensation; Represents the forecast confidence, which is obtained by evaluating the accuracy of historical forecasts and is calculated as ,in Indicates the predicted value of the previous moment for the current moment; is the confidence threshold, which is set to 0.7, indicating that when the prediction confidence is lower than 0.7, the weight of the feedforward compensation begins to decrease.

[0241] In outdoor environments with complex wind disturbances, such as high-rise building exterior wall inspection scenarios, the signal peak real-time detection and feedforward compensation module can predict disturbance changes caused by gusts, generate compensation control quantities in advance, reduce the impact of wind disturbances on system stability, and ensure that the drone can hover stably at the inspection position and the robotic arm can operate accurately.

[0242] like Figure 9 As shown, the multi-module collaborative dynamic fusion module 7 includes a basic control unit 71, a weight calculation unit 72, a fusion processing unit 73 and an output limitation unit 74.

[0243] The basic control unit 71 is used to generate a basic control variable based on the system error vector and historical control input. The basic control variable is generated based on the time delay estimation (TDE) principle:

[0244] ,

[0245] in, represents the basic control quantity vector; Indicates the basic PD control quantity, represents the total disturbance compensation estimated by the disturbance observer; and are the proportional and differential gain matrices respectively; and represent the error vector and error change rate vector respectively.

[0246] Disturbance Observer Design:

[0247] ,

[0248] in, represents the disturbance estimate at the previous moment; is the observer gain matrix, and the diagonal elements are set to 0.7; represents the control input at the previous moment; Indicates the PC control quantity at the previous moment; represents the nominal mass matrix of the system, which is a fixed diagonal matrix containing only the estimated values ​​of the basic inertial parameters; and Represent the velocity vectors at the current and previous moments respectively; is the sampling period, .

[0249] The weight calculation unit 72 is in communication with the basic control unit 71 and is used to calculate the fusion weight of the basic control variable, the sliding mode control variable and the feedforward compensation control variable based on the size and change rate of the system error vector. The weight calculation formula is:

[0250] ,

[0251] ,

[0252] in, represents the weight coefficient of sliding mode control; is the baseline weight, set to 0.6; is the adjustment item based on the error size, and the calculation formula is ,in For the reference error size, set it to 0.1 (for angle units, the unit is radians, for position units, the unit is meters); Represents the weight coefficient of feedforward compensation; is the baseline weight, set to 0.8; is an exponential decay function based on the error size; is the regularization parameter, set to 0.05, which controls the rate at which the weight changes with the error.

[0253] This dynamic weight adjustment strategy ensures the optimal control performance of the system under different operating conditions: when the error is large, the weight of sliding mode control is increased to improve the system convergence speed; when the error is small, the weight of feedforward compensation is increased to improve the system's ability to predict and compensate for disturbances.

[0254] The fusion processing unit 73 is in communication with the weight calculation unit 72 and is used to synthesize the final control instruction according to the fusion weight. The fusion formula is:

[0255] ,

[0256] in, represents the final control instruction vector; is the TDE basic control quantity vector; is the sliding mode control quantity vector; is the feedforward compensation control quantity vector.

[0257] The output limiting unit 74 is in communication with the fusion processing unit 73 to ensure that the final control instruction is within the actuator limit range. The limiting function is:

[0258] ,

[0259] in, represents the i-th control quantity component after limitation; represents the i-th control quantity component; and The upper and lower limits of the i-th control variable are determined based on the physical characteristics of the actuator. For example, for rotor thrust control, the limit range is typically 0% to 100% throttle; for manipulator joint control, the limit range is the maximum torque value of the motor, typically between 1 and 5 Nm.

[0260] In addition, to avoid sudden changes in the control quantity, a change rate limit is set:

[0261] ,

[0262] in, The maximum allowable rate of change of the i-th control variable component is set based on the system's dynamic response characteristics. For rotor thrust control, the maximum rate of change is typically set to 20% / 100ms (i.e., a maximum throttle change of 20% per 100ms); for manipulator joint control, the maximum rate of change is typically set to 1Nm / 100ms.

[0263] In the complex scenarios of power tower inspection and maintenance, the multi-module collaborative dynamic fusion module can dynamically adjust control strategies based on different mission phases and environmental conditions. For example, during long-distance flight, the system prioritizes stability for smoother control; when approaching the inspection target, the system prioritizes precise positioning for more accurate control; and during robotic arm operation, the system prioritizes anti-disturbance, ensuring that robotic arm operations do not affect the stability of the flight platform.

[0264] like Figure 10 As shown, the execution drive module 8 includes a control distribution unit 81 , a rotor drive unit 82 , a robotic arm drive unit 83 and a safety monitoring unit 84 .

[0265] The control distribution unit 81 is used to decompose the final control command into rotor thrust control signals, rotor direction control signals and manipulator joint control signals. The distribution of rotor control signals is based on the quadrotor dynamics model:

[0266] ,

[0267] in, Indicates the total thrust in Newton (N); 、 and Represents the rolling, pitching and yaw moments respectively, with the unit being Newton-meter (N Indicates the thrust of the four rotors, in Newton (N); It is the distribution matrix, which is related to the rotor layout and rotation direction. For a standard X-type quadrotor, the B matrix is ​​usually:

[0268] ,

[0269] in, Indicates the distance from the rotor to the center, in meters (m), which is 0.25m in this system; The conversion factor from thrust to torque is expressed in meters (m), which is related to the rotor characteristics and is approximately 0.01m in this system. The reverse distribution is:

[0270] ,

[0271] in, is the inverse of the distribution matrix, which is used to calculate the thrust of each rotor from the desired total thrust and torque.

[0272] For vectoring rotors, the rotor pitch angle must also be assigned to achieve the desired thrust vectoring:

[0273] ,

[0274] in, Indicates the tilt angle of the four rotors in radians (rad); represents the vector distribution function, which converts the desired force and torque into the rotor tilt angle based on the principles of mechanics. In this system, the rotor can be tilted within a range of ±15° to achieve horizontal thrust vector control.

[0275] The distribution of the robot arm joint control signals is based on the kinematic and dynamic models of the robot arm, which converts the desired posture of the end effector into the angle and torque control signals of each joint.

[0276] The rotor drive unit 82 is in communication with the control distribution unit 81 and is used to execute the rotor thrust control signal and the rotor direction control signal. The rotor drive uses a PWM signal to control the motor speed. The relationship between speed and thrust is:

[0277] ,

[0278] in, Represents the thrust of the i-th rotor, in Newtons (N); is the thrust coefficient, in N·s , which is related to the rotor characteristics and is approximately N Indicates the angular velocity of the rotor in radians per second (rad / s). The rotor direction is controlled by a digital servo, and the servo angle is adjusted by controlling the duty cycle of the PWM signal:

[0279] ,

[0280] in, Indicates the rotor tilt angle in radians (rad); is the angle coefficient, the unit is rad / %, in this system it is about 0.00436rad / % (equivalent to ; Indicates the duty cycle of the PWM signal in percentage , usually ranging from 5% to 10%; Indicates the duty cycle of the neutral position, typically 7.5%.

[0281] The manipulator drive unit 83 is in communication with the control distribution unit 81 and is responsible for executing control signals for the manipulator joints. The manipulator drive utilizes a serial servo network, sending control commands via a digital bus to achieve closed-loop control of position, velocity, and torque. In this system, the communication frequency of the manipulator joint motors is 50 Hz, with a position control accuracy of 0.1° and a torque control accuracy of 0.01 Nm.

[0282] The safety monitoring unit 84 is in communication with the rotor drive unit 82 and the manipulator drive unit 83 to monitor the actuator response status and activate the protection mechanism in abnormal situations. Safety monitoring includes:

[0283] Motor temperature monitoring: When the temperature exceeds 80°C, the motor power output is limited

[0284] Battery voltage monitoring: When the battery voltage is lower than the minimum safety value (3.5V per cell), low battery protection is triggered

[0285] Actuator response monitoring: When the actuator response deviates from the expected value by more than a threshold (±10% for rotor speed, ±5° for servo angle, and ±2° for robot arm joints), fault detection and processing are initiated

[0286] Abnormal posture monitoring: When the posture angle exceeds the safety range (±45°), the automatic stabilization program is triggered

[0287] Safety monitoring is particularly important in high-altitude power line inspection scenarios. For example, when the system approaches high-voltage power lines, the safety monitoring unit can detect potential electromagnetic interference affecting sensors and actuators, promptly initiating protective measures to ensure system safety. During long missions, the safety monitoring unit can monitor battery status and motor temperature, alerting the system to return home before energy is exhausted or overheated, preventing accidental falls or damage.

[0288] The system also includes a communication interface module 9 and a task planning module 10 for receiving external control instructions and task parameters, and planning the system motion trajectory and the robot arm operation sequence.

[0289] The communication interface module 9 utilizes 2.4GHz wireless communication technology, supporting real-time data transmission with a range of up to 1000 meters and a data rate of 10Mbps. The data frame structure consists of a header (2 bytes), a command type (1 byte), a data payload (variable length, up to 256 bytes), and a checksum (2 bytes), ensuring reliable and real-time communication. Communication utilizes a TDMA (time division multiple access) mechanism, with a system status reporting frequency of 50Hz and a control command issuance frequency of 20Hz, ensuring real-time and stable control.

[0290] The task planning module 10 plans the system's motion trajectory and the sequence of manipulator operations based on received external control commands, task parameters, and system state estimates. Trajectory planning utilizes the principle of minimum energy consumption to generate a smooth third-order spline curve, ensuring the continuity and efficiency of system motion. Trajectory points are generated at a frequency of 20 Hz, matching the 100 Hz control loop. The controller updates the desired trajectory points every five cycles.

[0291] In power inspection scenarios, the task planning module can plan the optimal inspection route based on the distribution of power equipment, reducing energy consumption. It also plans the robotic arm's operation sequence based on the inspection requirements of different equipment to ensure comprehensiveness and accuracy. For example, insulator inspection requires the robotic arm to capture images from multiple angles, while bolt tightening inspection requires the robotic arm to contact the bolts and measure torque. Based on these requirements, the task planning module automatically generates the corresponding operation sequence, reducing the operator's burden and improving work efficiency.

[0292] The lightweight quadrotor vector drone-robotic arm system of the present invention has been experimentally verified in scenarios such as power line inspection and maintenance, and building exterior wall inspection and maintenance. Attitude stability tests under different load conditions showed that:

[0293] 1. When the robot arm is unloaded, the system's attitude stability (root mean square of angular deviation) is approximately ±0.5°, and its position stability is approximately ±5cm.

[0294] 2. With a robotic arm load of 500g (approximately 1 / 3 of the system's own weight), the system's attitude deviation is controlled within ±1.5° and its position deviation is controlled within ±8cm using the TDE-SMC control algorithm of this invention.

[0295] 3. Compared with the traditional PID control algorithm, the control algorithm of the present invention improves the posture stability by 70% and the position stability by 65%. The advantages are more obvious especially in the cases of rapid movement of the robotic arm and sudden load changes.

[0296] 4. The system has a success rate of over 90% when performing delicate operations (such as tightening bolts, plugging and unplugging connectors, etc.), significantly improving the application capabilities of drone systems in physical interaction scenarios.

[0297] Through optimized control algorithms and system architecture design, the present invention successfully solves the problem of dynamic coupling between vector drones and robotic arm systems, realizes a coherent detection-execution operation process, and expands the application boundaries of drones in industrial inspection, emergency rescue, urban services and other fields.

[0298] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

Claims

1. Lightweight quadrotor vector drone-robotic arm system, characterized by: include: Sensor perception module, used for: Collect UAV attitude data and position data; Collect robot arm joint status data; A state estimation module, in communication with the sensing module, is configured to: Receiving posture data, position data and joint status data sent by the sensing perception module; estimating a system state and calculating an error vector based on the posture data, the position data, and the joint state data; An adaptive delay parameter optimization module, in communication with the state estimation module, is configured to: receiving an error vector sent by the state estimation module; Dynamically adjusting the delay parameter based on the changing characteristics of the error vector; A multi-level disturbance decoupling and estimation module is in communication with the state estimation module and the adaptive delay parameter optimization module, and is used to: receiving the error vector sent by the state estimation module and the delay parameter sent by the adaptive delay parameter optimization module; Decomposing the system disturbance into low-frequency disturbance, medium-frequency disturbance and high-frequency disturbance according to the frequency characteristics of the error vector; Based on the time delay parameter, respectively estimating the low-frequency disturbance, the medium-frequency disturbance and the high-frequency disturbance; The sliding mode boundary layer adaptive control module is in communication with the state estimation module and the multi-level disturbance decoupling and estimation module, and is used to: Receiving the error vector sent by the state estimation module and the disturbance estimation result sent by the multi-level disturbance decoupling and estimation module; constructing a sliding surface based on the error vector; Dynamically adjusting the boundary layer thickness according to the magnitude of the disturbance estimation result; generating a sliding mode control variable based on the sliding mode surface and the boundary layer thickness; The signal peak real-time detection and feedforward compensation module is in communication with the state estimation module and the multi-level disturbance decoupling and estimation module, and is used to: Receiving the error vector sent by the state estimation module and the disturbance estimation result sent by the multi-level disturbance decoupling and estimation module; Detecting a change trend of the error vector to determine whether it exceeds a preset threshold; When the variation trend of the error vector exceeds the preset threshold, predicting future disturbances and generating a feedforward compensation control amount; The multi-module collaborative dynamic fusion module is in communication with the state estimation module, the multi-level disturbance decoupling and estimation module, the sliding mode boundary layer adaptive control module, and the signal peak real-time detection and feedforward compensation module, and is used to: Receive the error vector sent by the state estimation module, the disturbance estimation result sent by the multi-level disturbance decoupling and estimation module, the sliding mode control amount sent by the sliding mode boundary layer adaptive control module, and the feedforward compensation control amount sent by the signal peak real-time detection and feedforward compensation module; generating a basic control variable based on the error vector; Dynamically adjusting the weights of the basic control variable, the sliding mode control variable, and the feedforward compensation control variable according to the size and change trend of the error vector; generating a final control instruction based on the weight; An execution driving module is communicatively connected with the multi-module collaborative dynamic fusion module and is used to: Receiving the final control instruction sent by the multi-module collaborative dynamic fusion module; Decomposing the final control instruction into a rotor drive signal and a manipulator joint control signal; Execute the rotor drive signal and the robotic arm joint control signal.

2. The system according to claim 1, wherein: The sensing module includes: Attitude sensing unit, used to collect body angle, angular velocity and acceleration data; Position sensing unit, used to collect body position and speed data; Force feedback sensing unit, used to collect the joint torque and position data of the robotic arm; A data preprocessing unit is communicatively connected with the posture sensing unit, the position sensing unit and the force feedback sensing unit, and is used for performing noise filtering and temperature compensation on the collected raw data.

3. The system according to claim 1, wherein: The state estimation module includes: A data fusion unit, configured to fuse the posture data, position data, and joint state data sent by the sensing module to generate a system state estimation result; an error calculation unit, communicatively connected to the data fusion unit, for comparing the system state estimation result with the expected trajectory and calculating an error vector and its rate of change; The system state estimation result includes the body posture, position and robot arm joint state, and the error vector includes posture error, position error and joint error.

4. The system according to claim 1, wherein: The adaptive delay parameter optimization module includes: a feature extraction unit, configured to calculate a time series characteristic of a state quantity based on the error vector; a correlation analysis unit, connected in communication with the feature extraction unit, for evaluating the degree of correlation of state quantities under different delay values; a parameter search unit, communicatively connected to the correlation analysis unit, and configured to search for an optimal delay parameter within a preset delay range; A smooth transition unit is communicatively connected to the parameter search unit and is used to perform smoothing processing on the optimal delay parameter to avoid sudden changes in the delay parameter.

5. The system according to claim 1, wherein: The multi-level disturbance decoupling and estimation module includes: Frequency domain analysis unit, used to perform spectrum analysis on system disturbances and decompose disturbance signals into different frequency bands; a low-frequency disturbance estimation unit, communicatively connected to the frequency domain analysis unit, for processing structural disturbances such as changes in the load of the manipulator; An intermediate frequency disturbance estimation unit, which is in communication with the frequency domain analysis unit and is used to process environmental disturbances such as wind and airflow; a high-frequency disturbance estimation unit, communicatively connected to the frequency domain analysis unit, for processing measurement disturbances such as sensor noise; The disturbance reconstruction unit is in communication with the low-frequency disturbance estimation unit, the medium-frequency disturbance estimation unit and the high-frequency disturbance estimation unit, and is used to integrate the disturbance estimation results of each frequency band to generate a complete disturbance estimation.

6. The system according to claim 1, wherein: The sliding mode boundary layer adaptive control module includes: a sliding surface calculation unit, configured to construct a sliding surface based on the error vector and its rate of change; a boundary layer evaluation unit, in communication with the sliding surface calculation unit, and configured to evaluate the boundary layer thickness required by the current system according to the magnitude of the disturbance estimation result; a thickness adjustment unit, in communication with the boundary layer evaluation unit, for dynamically adjusting boundary layer parameters, increasing thickness when the disturbance is large, and decreasing thickness when the disturbance is small; A control amount generating unit is in communication with the sliding surface calculating unit and the thickness adjusting unit, and is used to generate a smooth sliding mode control instruction through a saturation function.

7. The system according to claim 1, wherein: The signal peak real-time detection and feedforward compensation module includes: a feature extraction unit, configured to detect a changing trend of the error vector; a threshold determination unit, communicatively connected to the feature extraction unit, and configured to determine a trigger threshold based on statistical characteristics of the error vector; a disturbance prediction unit, communicatively connected to the threshold judgment unit, configured to predict disturbance changes in a short period of time in the future when the change trend of the error vector exceeds the trigger threshold; The feedforward generating unit is in communication with the disturbance predicting unit and is used to generate a feedforward compensation control amount based on the predicted disturbance change.

8. The system according to claim 1, wherein: The multi-module collaborative dynamic fusion module includes: A basic control unit, configured to generate a basic control variable based on the error vector and historical control inputs; a weight calculation unit, communicatively connected to the basic control unit, for calculating a fusion weight of the basic control amount, the sliding mode control amount, and the feedforward compensation control amount according to the magnitude and change rate of the error vector; a fusion processing unit, communicatively connected to the weight calculation unit, for synthesizing a final control instruction according to the fusion weight; An output limiting unit is communicatively connected to the fusion processing unit and is used to ensure that the final control instruction is within the actuator limitation range.

9. The system according to claim 1, wherein: The execution drive module includes: a control distribution unit, configured to decompose the final control instruction into a rotor thrust control signal, a rotor direction control signal, and a manipulator joint control signal; a rotor drive unit, communicatively connected to the control distribution unit, for executing the rotor thrust control signal and the rotor direction control signal; a robotic arm drive unit, communicatively connected to the control distribution unit, and configured to execute the robotic arm joint control signals; A safety monitoring unit is communicatively connected with the rotor drive unit and the manipulator drive unit, and is used to monitor the actuator response status and activate a protection mechanism in abnormal situations.

10. The system according to claim 1, wherein: The system further comprises: The communication interface module is communicatively connected to the multi-module collaborative dynamic fusion module and is used to: Receive external control instructions and task parameters; Send system status information and task execution results to external devices; A mission planning module, in communication with the communication interface module and the state estimation module, is configured to: Receiving external control instructions and task parameters sent by the communication interface module; Receiving a system state estimation result sent by the state estimation module; Planning the system motion trajectory and the manipulator operation sequence based on the external control instructions, the task parameters and the system state estimation result; The system motion trajectory is sent to the state estimation module as a desired trajectory.

Images