Control method of cleaning robot for unmanned aerial vehicle to autonomously recover photovoltaic modules

Abstract

The invention provides a control method for an unmanned aerial vehicle to autonomously recover a photovoltaic module cleaning robot, and the method comprises the steps: firstly receiving photovoltaic module position information sent by a ground base station by the unmanned aerial vehicle, controlling the unmanned aerial vehicle to fly to a target area through a position loop, and hovering at a set height; then, a visual system identifies an ArUco mark carried by the cleaning robot, and the position, rotation matrix and number information of the ArUco mark in the image coordinate system and the unmanned aerial vehicle body coordinate system are solved; and after coordinate transformation, attitude data in the ENU coordinate system are obtained. Based on a multi-mode fuzzy PID control strategy, the speed of the unmanned aerial vehicle is dynamically adjusted, and real-time tracking and relative position maintaining of the cleaning robot are achieved. And the airborne computer and the mechanical claw system perform cross-domain cooperative control, drive the mechanical claw to execute the grabbing action and judge whether grabbing succeeds or not. And finally, the unmanned aerial vehicle transports the cleaning robot to a designated position according to a base station instruction. The method has good reliability and high operation efficiency in unmanned aerial vehicle autonomous recovery of the photovoltaic module cleaning robot.

Description

TECHNICAL FIELD

[0001] The application relates to a control method for autonomously recycling a photovoltaic component cleaning robot by a UAV, and belongs to the technical field of photovoltaic operation and maintenance. BACKGROUND

[0002] As an important part of clean energy, the energy conversion efficiency of photovoltaic power generation is closely related to the surface cleaning degree of photovoltaic components. In order to ensure the efficient power generation performance of photovoltaic components, automatic cleaning robots have been widely used in the daily cleaning of photovoltaic components. However, due to the limitations of mechanical structure and power drive, the existing cleaning robots cannot cross the physical gap between adjacent photovoltaic component arrays, thereby limiting their cleaning efficiency in large-scale photovoltaic power stations, and they can usually only complete the cleaning task within a single photovoltaic component or a single array.

[0003] In order to overcome the above limitations, the industry has proposed a scheme of using a UAV as a carrying platform to carry out cross-row scheduling and transportation of cleaning robots. Although this scheme is theoretically feasible, there are still many key technical bottlenecks in actual application, especially the full autonomous recycling method of the UAV is basically in the blank field. There are mainly two recycling methods at present: one is manual on-site recycling, which requires the operator to reach the work position for manual grabbing, and has the problems of low work efficiency and poor operation safety; the other is to use a remote control pilot to operate the UAV to grab, which can avoid direct climbing of personnel, but still belongs to semi-automation, has the problems of high skill requirement for the operator and difficult accurate control in the grabbing process, and cannot realize full automation.

[0004] There are several technical problems to be solved in applying UAV grabbing technology to the recycling of photovoltaic component cleaning robots. First, the positioning accuracy is insufficient: when working on a large-scale photovoltaic array, the UAV needs to balance between the working radius and the positioning accuracy, and the global navigation satellite system alone cannot meet the centimeter-level accuracy required for end grabbing. Second, the dynamic tracking stability is insufficient: due to external wind disturbance and UAV attitude fluctuation, there is relative motion between the body and the cleaning robot in the hovering state, and traditional control methods cannot realize accurate pose keeping under high dynamic conditions, reducing the success rate of grabbing. In addition, the cross-domain cooperative control mechanism is missing: the flight control system and the mechanical claw execution system of the UAV are usually operated independently by different control units, and there is no unified decision center for cooperative control of flight attitude and grabbing action, and a complete "perception-decision-execution" closed loop cannot be formed.

[0005] Therefore, a method capable of realizing full autonomy, high precision and high reliability of the unmanned aerial vehicle for the photovoltaic module cleaning robot is proposed, which has important practical significance. The method can not only improve the automation level of photovoltaic power station operation and maintenance, but also promote the intelligent development of the photovoltaic industry, meet the cleaning needs of large-scale photovoltaic power stations, and improve the overall efficiency of photovoltaic power generation. SUMMARY

[0006] Technical problem: Aiming at solving the problem that small cleaning robots cannot work across rows on photovoltaic panels, and the current unmanned aerial vehicle recovery method is not mature, resulting in the unmanned aerial vehicle cannot accurately recover and transport the cleaning robot autonomously. The existing method relies on manual transportation and transfer or pilot operation of the unmanned aerial vehicle for recovery, which cannot realize the automation and efficient operation required by engineering, resulting in serious constraints on the continuity of photovoltaic operation and maintenance.

[0007] Technical solution: In order to achieve the above purpose, the present application provides the following technical solution:

[0008] The control method for the unmanned aerial vehicle to autonomously recover the photovoltaic module cleaning robot of the present application comprises the following steps in sequence:

[0009] S1: The inspection unmanned aerial vehicle scans the photovoltaic array through the onboard vision system, identifies the dirty photovoltaic panel and clean photovoltaic panel number and its GNSS coordinates, and detects whether there is an idle cleaning robot on the clean panel. The inspection unmanned aerial vehicle sends a cleaning demand data packet to the ground base station, which contains the position information of the photovoltaic panel to be cleaned and the idle robot. The transport unmanned aerial vehicle flies to the target recovery point based on the GNSS and IMU sensor data, using the PID position loop control algorithm. The unmanned aerial vehicle accurately arrives above the target photovoltaic panel and maintains stable hovering at a specified height, laying a foundation for subsequent visual recognition and grabbing operation.

[0010] S2: Before performing the recovery task, the camera of the onboard vision system needs to be calibrated to ensure the accuracy of ArUco code image recognition and position calculation. Calculate the camera intrinsic parameter matrix and distortion coefficient to complete the intrinsic parameter calibration. Then, taking the take-off point as the origin of the ENU coordinate system, solve the rotation matrix and translation vector of the camera relative to the coordinate system to complete the extrinsic parameter calibration. Finally, according to the installation angle of the photovoltaic module, plan the flight height and path of the unmanned aerial vehicle, and make the camera lens perpendicular to the ground to optimize the shooting effect and identify the ArUco code in real time, ensuring accurate positioning support for subsequent cleaning robot recovery operation.

[0011] S3: The calibrated camera processes the real-time collected image, recognizes the ArUco code carried on the top of the cleaning robot, calculates the relative pose of the target in the camera coordinate system by PnP algorithm, thereby obtains the accurate pose of the cleaning robot relative to the camera, and converts the pose to the UAV body coordinate system according to the fixed installation relationship between the camera and the UAV body, transforms the relative pose in the body coordinate system to the ENU global coordinate system by fusing the GNSS position and IMU attitude data of the UAV itself, outputs the global absolute position and orientation of the robot, and completes the accurate positioning.

[0012] S4: The airborne computer calculates the position error of the cleaning robot relative to the expected hovering point according to the relative position of the cleaning robot in the body coordinate system output by the vision system; the system compares the error norm with the preset threshold, autonomously judges and switches the control mode; the controller takes the normalized position error E and its change rate DE as input, adjusts the PID parameters online through fuzzy reasoning. Finally, the system calculates the control command according to the current mode and the adjusted parameters, and sends it to the flight control system to drive the UAV to realize fast and stable adaptive tracking and position keeping.

[0013] S5: The airborne computer continuously monitors the relative pose of the UAV-robot, and performs the grasping and positioning judgment. When the spatial pose error is continuously and stably within the threshold, and the IMU data shows that the body shaking is below the threshold, the control system authorizes to enter the grasping stage. In this stage, the decision layer generates a smooth descending command, which is accurately executed by the flight control bottom layer PID. The outer loop decision layer synchronously performs safety monitoring, and if the error is out of limit, it will immediately interrupt the descent and trigger the emergency return. When approaching the target, the electromagnet at the bottom of the mechanical claw is pre-started, and its magnetic field guides the claw head to complete passive compliant positioning. When the loop current rises significantly due to adsorption, the system determines that the contact is successful, and then instructs the mechanical claw to close to ensure reliable connection.

[0014] S6: According to the overall task flow, specify comprehensive evaluation indexes to evaluate the feasibility, reliability and efficiency of the scheme. Three evaluation indexes are given in the invention: recovery success rate, positioning absolute accuracy and relative accuracy, grasping stability and safety. The overall and local steps are comprehensively and objectively evaluated, which verifies the scientificity of the recovery process and provides quantitative reference for subsequent directional improvement.

[0015] Further, in step S1, the UAV receives the target photovoltaic panel position information sent by the ground base station and controls to fly above the target photovoltaic panel through the position loop, and the specific steps include:

[0016] S1.1: Task information perception and reporting: The inspection unmanned aerial vehicle conducts comprehensive inspection on the photovoltaic array with the aid of the on-board visual system. The system automatically identifies the number information of dirty and clean photovoltaic panels and synchronously records their GNSS positioning coordinates. At the same time, it detects whether there is a cleaning robot in working condition on the surface of the clean photovoltaic panel. After completing data collection, the inspection unmanned aerial vehicle transmits the cleaning demand data packet to the ground base station through the wireless data link, which clearly marks the location information of the dirty photovoltaic components to be cleaned and the location information of the clean photovoltaic components where the idle cleaning robot exists.

[0017] S1.2: Task planning and instruction issuing: The ground base station receives the inspection data and uses the maximum weight matching of bipartite graph for task planning. The base station calculates the optimal recovery and deployment path according to the photovoltaic panel distribution and robot position information, generates flight task instructions containing target recovery and deployment positions, and issues the instructions to the designated transport unmanned aerial vehicle through the wireless communication link.

[0018] S1.3: Unmanned aerial vehicle autonomous flight and hovering: The transport unmanned aerial vehicle receives the base station instructions, obtains real-time position and attitude data through the built-in GNSS receiver and inertial measurement unit (IMU). It uses the PID position loop control algorithm to generate three-dimensional space motion instructions (position and speed control). The unmanned aerial vehicle first flies to the target recovery point, accurately arrives above the target photovoltaic panel through closed-loop control, and maintains stable hovering at the specified height, creating favorable conditions for subsequent visual recognition and grabbing operations.

[0019] Further, the step S2 is for the subsequent ArUco code image recognition and position solving, and the camera internal and external parameter calibration used by the visual system, the shooting posture planning adjustment, including the following steps:

[0020] S2.1: Camera internal parameter calibration: A checkerboard calibration board is used to calculate and obtain the internal parameter matrix and lens distortion coefficient of the camera through multiple angle image acquisition, and the internal parameter calibration of the visual system is completed. Specifically, first, prepare a checkerboard calibration board with known size, and use each corner point on the calibration board as a reference in the 3D world coordinate system; take multiple images of the calibration board; then detect and optimize the calibration board corner points; then perform 3D-2D point matching; finally, calculate the internal parameter matrix and distortion coefficient of the camera through the calibration algorithm.

[0021] S2.2: External parameter coordinate system establishment: Establish an ENU coordinate system with the take-off point of the unmanned aerial vehicle as the origin, and solve the external parameters of the camera relative to this coordinate system, including the rotation matrix R and the translation vector T, to determine the attitude and position of the camera in the global space. Specifically: take the take-off point as the origin of the ENU coordinate system, calculate the rotation matrix of the body coordinate system relative to the ENU coordinate system according to the attitude angle of the unmanned aerial vehicle; further, calculate the rotation matrix of the gimbal coordinate system relative to the unmanned aerial vehicle coordinate system according to the attitude angle of the gimbal relative to the unmanned aerial vehicle; obtain the final rotation matrix of the camera coordinate system relative to the ENU coordinate system through these two rotation matrices; then, calculate the translation T through the translation of the camera relative to the body and the position of the body in the ENU. Finally, the external parameters of the camera relative to the ENU coordinate system are obtained, including the rotation matrix (R) and the translation vector (T), thus describing the position and direction of the camera in the ENU coordinate system.

[0022] S2.3: Shooting attitude planning and adjustment: Combine the actual installation angle of the photovoltaic module to plan the flight height and path of the unmanned aerial vehicle, and control the gimbal mechanism to make the camera optical axis approximately perpendicular to the ground. Specifically: adjust the flight height and path of the transport unmanned aerial vehicle according to the installation angle of the photovoltaic module, optimize the camera shooting effect, and ensure that the gimbal camera lens is perpendicular to the ground, and perform real-time ArUco code recognition and positioning.

[0023] Further, in step S3, the three-dimensional coordinates of the cleaning robot in the body coordinate system and its global absolute position are obtained by solving the two-dimensional image information, and the complete information package containing the position data and the identity number is fed back to the ground base station. Including the following steps:

[0024] S3.1: ArUco code detection and information extraction: start the on-board vision system, process the real-time collected images; identify the ArUco code carried on the top of the cleaning robot, extract the pixel coordinates of the corner points and read the unique number. Specifically: the transport unmanned aerial vehicle arrives above the target photovoltaic panel and hovers stably, and the on-board vision system is activated. The integrated recognition algorithm is used to quickly locate the ArUco code mark on the top of the cleaning robot. After successful recognition, the system extracts the pixel coordinates of the corner points and the identity number.

[0025] S3.2: Solve the relative pose in the camera coordinate system: use the internal parameter matrix and distortion coefficient obtained by pre-camera calibration, and combine the known physical size of the ArUco code, to solve the three-dimensional relative position and rotation matrix of the ArUco code in the camera coordinate system through the PnP algorithm, to obtain the accurate pose of the robot relative to the camera. Specifically: based on the perspective projection principle, the three-dimensional pose of the mark in the camera coordinate system is calculated according to the pre-calibrated camera parameters and the physical size of the mark.

[0026] S3.3: Coordinate Transformation to the Body Coordinate System: Based on the fixed installation relationship between the camera and the UAV body, the pose calculated in the camera coordinate system in S3.2 is transformed to the UAV body coordinate system using the extrinsic parameters of the camera-body connection. Specifically, the pose is transformed to the body coordinate system using the fixed installation parameters of the camera and the body, and then further transformed to the ENU global coordinate system with the takeoff point as the origin by fusing GNSS and IMU data, thereby obtaining the robot's global absolute position and orientation.

[0027] The drone's onboard computer sends a complete positioning data packet, containing the robot's ID, global coordinates, attitude, and timestamp, to the ground base station. The ground system then updates the robot's status and location accordingly, providing real-time data support for subsequent scheduling and path planning.

[0028] Furthermore, in step S4, based on the position information in the body coordinate system calculated in S3, a multimodal fuzzy PID speed control is proposed, enabling the transport drone to hover precisely above the cleaning robot. This includes the following steps:

[0029] S4.1: Modal determination: The UAV's onboard computer receives the relative position of the cleaning robot in the body coordinate system calculated by the onboard vision system in real time. ; Calculate the position error at the current moment ,in The desired hovering point for the drone; based on the norm of the horizontal error. The system compares its performance against preset thresholds, autonomously determining and switching control modes. When the drone is far from the target robot, the control system employs a position PID-dominated mode. This mode uses global position deviation as the primary control input to control the drone's flight trajectory, ensuring its rapid approach to the target area. In this mode, the system prioritizes global flight stability and anti-interference capabilities. As the distance between the drone and the target gradually decreases, the system smoothly switches to a speed PID-dominated mode. This mode uses relative speed as the core control variable, responding quickly to the robot's movements. This switching effectively improves the system's dynamic tracking accuracy at close range and reduces overshoot.

[0030] S4.2: Fuzzy PID Parameter Tuning: To overcome the shortcomings of traditional PID parameters being fixed and having poor adaptability in complex environments, a fuzzy logic controller is introduced for each control mode. This controller takes the normalized position error E and error change rate ΔE as inputs, performs inference through a predefined fuzzy rule set, and outputs the PID control parameters online in real time. The adjustment amount; this mechanism enables the control system to dynamically adjust its control strategy according to different tracking states, combining responsiveness and stability;

[0031] S4.3: Multimodal control law calculation and execution: Based on the currently active control mode and the fuzzy tuned PID parameters, the final control command is calculated. The generated control command is sent to the UAV's flight control system to drive the UAV to achieve dynamic tracking, thus completing stable tracking and position holding.

[0032] Furthermore, in step S5, a method for grasping control and status determination based on cross-domain collaboration is provided. This system achieves safe and reliable grasping operations through close cooperation between the onboard computer, flight controller, and robotic gripper terminal, and utilizes an electromagnetic feedback mechanism for real-time diagnosis of the grasping status. The method includes the following steps:

[0033] S5.1: Grasping Position Determination: The onboard computer continuously monitors the relative pose of the drone and robot from S4; it sets a grasping position threshold based on the horizontal error and attitude angle, and authorizes entry into the descent grasping phase when the conditions are met. Specifically: Before the grasping process starts, the system performs a dual safety check on the current state. The onboard computer continuously monitors the real-time relative pose between the drone and the cleaning robot. When both the horizontal position deviation and the attitude angle error are less than the preset threshold, the system initially determines that the spatial conditions are met. At the same time, the system analyzes the IMU inertial measurement data to confirm that the drone's own sway amplitude is in a low-level stable state. The above conditions must be maintained stably for a period of time (spatial conditions: the current relative pose error is continuously stable within the grasping position threshold for more than 2 seconds; dynamic conditions: the sway amplitude of the drone calculated from the IMU data is lower than the stability threshold) before the system finally authorizes entry into the grasping phase.

[0034] S5.2: Altitude Descent and Safety Monitoring under Cascaded Control: The decision-making layer generates a smooth vertical descent speed command based on visual positioning results; the execution layer, controlled by the UAV flight controller, precisely executes the altitude descent through the underlying PID controller; simultaneously, the outer-loop decision-making layer continuously monitors safety: real-time detection of horizontal position error and attitude stability; if the error exceeds the safety threshold or stability is insufficient, the descent is immediately interrupted, triggering an emergency ascent command to return the UAV to a safe altitude and re-perform the grasping and positioning determination, thereby ensuring operational safety under external interference. Specifically: After entering the grasping phase, the system uses a cascaded control architecture to execute the altitude descent. The decision-making layer generates a smooth vertical descent command based on visual positioning results, and the execution layer precisely controls the altitude through the underlying flight controller. During this process, the outer-loop monitoring system remains active, continuously evaluating horizontal positioning accuracy and aircraft stability. Once a deviation exceeding the safe range or abnormal shaking of the aircraft is detected, the system immediately terminates the descent process, triggering an emergency ascent command to return the UAV to a safe altitude to await subsequent commands.

[0035] S5.3: Electromagnetic Induction Trigger and Grasping Execution: When the drone descends close to the cleaning robot, the onboard computer pre-activates the electromagnet at the bottom of the mechanical claw, generating a magnetic field. This magnetic field creates a lateral correction force as the drone approaches and grasps the metal object, guiding the claw to precise alignment and achieving passive, compliant positioning. Once the electromagnet attracts the object, the circuit current increases significantly. Based on this, the onboard computer determines that the attraction is successful and immediately instructs the claw to close, ensuring a reliable transport connection.

[0036] Furthermore, in step S6, representative evaluation indicators are selected for the overall recycling process to make a comprehensive evaluation. This includes the following steps:

[0037] The system automatically generates a structured runtime log after each task execution, recording key parameters for the entire sequence from identification and localization to final capture. The evaluation system mainly revolves around three core dimensions:

[0038] Recovery success rate: Statistics are based on a large number of repetitive trials. The system records the frequency of successful capture and improved transfer, compares this number with the total number of attempts, and calculates the baseline recovery success rate.

[0039] Absolute and relative accuracy of positioning: Absolute accuracy is determined by comparing the deviation between the robot's global coordinates calculated by vision and the high-precision differential GNSS measurement reference value; relative accuracy is determined by analyzing the fluctuation range and convergence characteristics of the relative pose data of the UAV and the cleaning robot during the hovering phase. This evaluation system provides a stable reference for grasping operations.

[0040] Grasping stability and safety: The Euler angles of the transported object are measured by the IMU measurement module to determine the stability of the gripping; the maximum axial force of the mechanical gripper is measured to determine the upper limit of the weight it can bear and to evaluate the safety of the gripping process.

[0041] Beneficial effects:

[0042] Compared with existing technologies, the control method for a drone-based autonomous photovoltaic module cleaning robot proposed in this invention has the following advantages:

[0043] (1) Precise Grasping Based on High-Precision ArUco Visual Positioning: The GNSS / IMU data of the UAV is deeply fused with the ArUco code information identified by the airborne vision system. Through precise camera calibration and multi-level transformation from camera to body and then to the global (ENU) coordinate system, centimeter-level precise positioning of the cleaning robot is achieved. This high-precision positioning lays the foundation for all subsequent autonomous operations, ensuring that the UAV can accurately identify and aim at the target in complex environments, thereby improving the success rate of autonomous grasping.

[0044] (2) Multimodal fuzzy PID control for highly adaptable and stable tracking: This invention proposes an innovative multimodal fuzzy PID control method. This method dynamically switches between position and velocity control based on the distance between the UAV and the target. The PID parameters are tuned in real-time by a fuzzy logic controller, maintaining high-precision and high-stability dynamic tracking and hovering of the cleaning robot. This method solves the problems of fixed parameters and poor environmental adaptability in traditional PID control, ensuring reliability during the grasping process.

[0045] (3) Cross-domain collaborative control mechanism for safe and reliable autonomous grasping: This invention constructs a cross-domain collaborative control closed loop between the airborne computer, flight controller, and mechanical gripper system. Through dual determination of spatial attitude and dynamic stability of the aircraft, the system can accurately grasp the grasping timing. During descent, a cascaded control and safety monitoring mechanism is adopted to ensure process safety. The innovative use of electromagnet induction for end-effector compliant positioning and contact status determination greatly increases the grasping success rate. This mechanism enables seamless connection between flight and grasping actions, ensuring high reliability and safety of the operation.

[0046] (4) Using drones for aerial transport to overcome the bottleneck of cleaning robots operating across rows: The core value of this invention lies in using drones as aerial transport platforms to enable cleaning robots to autonomously recover and transfer between different photovoltaic module arrays. This innovative technology directly solves the problem that existing cleaning robots cannot operate across physical gaps due to structural limitations, so that the operating range of cleaning robots is no longer limited to a single row, thereby breaking through the technical bottleneck of large-scale photovoltaic power plant promotion and application.

[0047] (5) Improve cleaning coverage and efficiency, and optimize photovoltaic operation and maintenance mode: By solving the problem of cross-row transfer, the cleaning robot can efficiently clean any photovoltaic module in the power station, which significantly improves the coverage and operation efficiency of automated cleaning. Compared with traditional manual cleaning or remote-controlled drone cleaning, the "cleaning robot cleaning + drone aerial transfer" mode proposed in this invention has outstanding performance in terms of cleaning effect, cost control and operation safety, and further optimizes the existing photovoltaic operation and maintenance mode.

[0048] (6) Constructing an air-ground collaborative system to enhance the intelligence level of photovoltaic power plant operation and maintenance: This invention integrates the originally isolated cleaning robot into a closed-loop, networked "air-ground collaborative" intelligent operation and maintenance system by using inspection drones to discover needs, ground base stations to intelligently plan, and transport drones to autonomously execute. This promotes the development of photovoltaic power plants from labor-intensive operation and maintenance to unmanned and intelligent operation and maintenance. Attached Figure Description

[0049] Figure 1 This is a flowchart of the method proposed in this invention;

[0050] Figure 2 Flowchart of steps for planning algorithms for transport drone missions;

[0051] Figure 3 A flowchart illustrating the steps involved in camera parameter calibration and shooting constraints.

[0052] Figure 4 A flowchart illustrating the steps involved in determining the relative position of the cleaning robot in the world coordinate system.

[0053] Figure 5 for The tuning rule table;

[0054] Figure 6 This is a flowchart of the steps in a multimodal fuzzy PID control algorithm.

[0055] Figure 7 RTK positioning ellipse error analysis diagram;

[0056] Figure 8 Response curves and corresponding data for drone positioning;

[0057] Figure 9 The response curve and corresponding data for drone tracking;

[0058] Figure 10 This refers to the change in Euler angles of the transported object during the transport process. Detailed Implementation

[0059] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. However, the described embodiments are only some embodiments of the present invention, and the protection scope of the present invention is not limited to the described embodiments.

[0060] S1: The ground base station sends the target photovoltaic panel location information to the drone through the task planning algorithm, and the drone flies to the target photovoltaic panel and hovers at the specified height through the position loop control;

[0061] Specifically, it includes the following steps ( Figure 2 ):

[0062] S11: After completing the scanning of an inspection unit or a specific area, the inspection drone structures the collected information and encapsulates it into a standard data packet. This packet includes the numbers of the dirty photovoltaic panels to be cleaned and their GNSS location sets, as well as the locations of clean photovoltaic panels where idle cleaning robots have been confirmed to exist, along with their corresponding numbers. The data packet is appended with a timestamp and the drone's identification identifier and is uploaded to the ground control base station via an encrypted wireless data link.

[0063] The S12 ground base station receives a set of location information for dirty photovoltaic panels reported by the inspection drone. and collection of location information of idle cleaning robots Then, the task planning process is initiated. This invention models the problem as an optimization problem with time and space constraints. The system possesses... The system uses transport drones, each capable of carrying only one cleaning robot at a time. The core task is to assign transport missions to these drones, with each mission... Defined as from the pickup point to unloading point The complete transportation process.

[0064] Introducing load weighting coefficient ( The cost per unit distance for loaded flight is higher than for unloaded flight. The objective function is defined as minimizing the maximum weighted flight time among all UAVs:

[0065]

[0066] in and Representing drones The set of empty and loaded flight segments, For the segment The estimated flight time.

[0067] The above objectives are achieved through a two-stage task planning algorithm.

[0068] (1) Task matching based on weighted cost:

[0069] Construct a weighted cost matrix Each element Representing the execution of tasks The estimated weighted time.

[0070]

[0071] in and These represent the unloaded and loaded cruise speeds of the drone, respectively. This problem involves calculating the geographical distance between two points. It is transformed into a minimum weighted matching problem in a bipartite graph, which is solved using the Hungarian algorithm. The output is a set of tasks. (in This minimizes the total weighted time cost of all allocations.

[0072] (2) Cluster path planning considering load balancing:

[0073] The tasks generated in the first phase are rationally assigned to K drones, and a task sequence is planned for each drone. For each task... Calculate its cost centroid :

[0074]

[0075] The centroid is geometrically biased towards the high-weight unloading point. This can more accurately reflect the time cost distribution of tasks. Using the K-means clustering algorithm, the cost centroids of all tasks are... Divided into There are two clusters. Tasks within each cluster are assigned to the same drone for execution.

[0076] S13: The UAV flight control system employs a position-loop PID control algorithm to generate spatial motion commands. The controller's input is the deviation between the target position and the UAV's current position. This deviation undergoes proportional, integral, and derivative calculations to output corresponding velocity and acceleration control commands. These commands are ultimately converted into control signals for the power system, driving the UAV towards the target point. The closed-loop control system operates continuously, overcoming trajectory deviations caused by external environmental factors such as wind disturbances, ensuring the stability and accuracy of the flight path.

[0077] S2: Calculate the camera intrinsic parameter matrix and distortion coefficients. Using the takeoff point as the origin of the ENU coordinate system, solve the camera's rotation matrix and translation vector relative to this coordinate system to complete the calibration of the camera's intrinsic and extrinsic parameters, preparing for subsequent ArUco image recognition and position calculation. Based on the installation angle of the photovoltaic modules, plan the UAV's flight altitude and path, and identify the ArUco code in real time.

[0078] Specifically, it includes the following steps ( Figure 3 ):

[0079] S21 Camera Intrinsic Calibration: The camera intrinsic parameter calibration section involves the camera's intrinsic parameter matrix. The calibration method involves calculating the distortion coefficients. The calibration is performed by photographing a chessboard of known dimensions from multiple angles. The key steps are as follows:

[0080] A checkerboard corner detector is used to detect corner points on a standard checkerboard pattern calibration board with squares of known size. A corner point accuracy enhancement module is used for sub-pixel-level optimization to improve detection accuracy. A 3D world coordinate generator method is used to generate the coordinates of the checkerboard corner points in a 3D world coordinate system, with the top-left corner of the checkerboard as the origin. Within the plane. Global optimization is performed using the camera intrinsic parameter calibration kernel to obtain the camera's intrinsic parameter matrix. And distortion coefficient. Formula shown:

[0081] Intrinsic parameter matrix Usually one The matrix represents the camera's focal length, principal point position, and other relevant parameters:

[0082]

[0083] in:

[0084] It is the focal length of the camera in the X and Y axis directions.

[0085] These are the coordinates of the camera's principal point (usually the image center).

[0086] It is the tangential distortion coefficient of the camera.

[0087] The distortion coefficients are represented by a vector containing radial and tangential distortions, as follows:

[0088]

[0089] in, It is the radial distortion coefficient. It is the tangential distortion coefficient.

[0090] S22 Camera External Parameter Calibration

[0091] The key steps in external parameter calibration are as follows:

[0092] 1. UAV attitude transformation: via Euler angles Convert the drone's attitude into a rotation matrix According to the ZYX Euler angle order:

[0093]

[0094] in:

[0095] It is around The matrix of axis rotation,

[0096] It is a matrix that rotates about the Y-axis.

[0097] It is a matrix that rotates about the X-axis.

[0098] 2. Gimbal Rotation Matrix: Calculates the rotation matrix from the gimbal camera to the body coordinate system. .

[0099] 3. Final rotation matrix calculation: Through matrix composition, the rotation matrix of the camera relative to the ENU coordinate system is obtained. :

[0100]

[0101] 4. Translation Vector Calculation: The camera's translation vector is given by the following formula:

[0102]

[0103] in, It is the displacement of the UAV relative to the ENU coordinate system. It refers to the displacement of the camera relative to the body of the machine.

[0104] S23 Real-time ArUco Code Recognition: Based on the installation angle of the photovoltaic modules, the flight altitude and path of the transport drone are adjusted to optimize the camera shooting effect, while ensuring that the gimbal camera lens is perpendicular to the ground and facing downwards, to perform real-time ArUco code recognition.

[0105] S3: Activate the onboard vision system, identify the ArUco code carried on the top of the cleaning robot, calculate its relative position, rotation matrix and robot number in the image coordinate system and the body coordinate system, and convert the pose information of the cleaning robot to the ENU coordinate system through coordinate transformation to obtain the precise position in the global coordinate system.

[0106] Specifically, it includes the following steps ( Figure 4 ):

[0107] S31: The airborne vision system processes images captured by the vision sensor in real time. The system performs feature detection on the input image, identifies and locates preset ArUco markers. After successful localization, the algorithm accurately calculates the pixel coordinates of the four corners of the marker, then decodes the binary matrix to obtain the unique ID code of the marker.

[0108] S32: Extract the image pixel coordinates of the four corner points of the ArUco code. Next, pose calculation from 2D image to 3D camera coordinates is performed. The pixel coordinates are then distorted using the intrinsic parameter matrix K and distortion coefficients obtained from camera calibration to obtain ideal coordinates in the normalized camera coordinate system. Subsequently, the following projection equations are established:

[0109]

[0110] in R is the scaling factor. Rotation matrix, t is Translation vector, This represents the homogeneous coordinates of the ArUco marker corner point in the marker coordinate system.

[0111] Since the actual physical dimensions of the ArUco marker are known, the system defines the marker plane as... This simplifies the equation to:

[0112]

[0113] in These are the first two columns of the rotation matrix. The overdetermined system of equations is solved iteratively using the correspondences of at least four corner points. The solution is optimized by minimizing the reprojection error during the solution process.

[0114]

[0115] in Represents the projection function. For the first ArUco coordinates of each corner point. The final calculated rotation matrix R and translation vector... This fully describes the rigid body transformation from the marker coordinate system to the camera coordinate system. The three-dimensional coordinates of the marker center in the camera coordinate system are given by... The orientation is given directly, while the rotation matrix is ​​characterized by converting it to Euler angles or preserving the matrix form. This transformation establishes a precise mapping from two-dimensional pixels to three-dimensional space.

[0116] S33: After obtaining the pose of the cleaning robot in the camera coordinate system via S32, the system performs a coordinate transformation to the body coordinate system. This spatial relationship is determined by a... homogeneous transformation matrix Full description:

[0117]

[0118] in, It is The identity orthogonal matrix defines the rotation relationship from the camera coordinate system to the body coordinate system; It is The vector represents the three-dimensional coordinates of the origin of the camera coordinate system in the body coordinate system.

[0119] Let the pose of the cleaning robot in the camera coordinate system be represented by a homogeneous transformation matrix. express:

[0120]

[0121] in and The pose of the cleaning robot in the body coordinate system is obtained by PnP calculation. Calculated via matrix multiplication:

[0122]

[0123] After unfolding, we get:

[0124]

[0125] Received Rotation matrix It describes the attitude of the marker coordinate system relative to the body coordinate system. Translation vector This gives the three-dimensional position of the origin of the marker in the body coordinate system.

[0126] S34: Calculate the pose of the cleaning robot in the body coordinate system. Next, the global navigation coordinate system is determined. This transformation fuses the relative pose information with the absolute geographic reference system, outputting the robot's absolute position and orientation in the northeast-sky coordinate system with the takeoff point as the origin.

[0127] The GNSS receiver provides the latitude, longitude, and altitude coordinates of the antenna phase center in the WGS-84 coordinate system. The IMU unit measures the three-axis attitude angles of the aircraft relative to inertial space and outputs the roll angle. , looking up ,course Parameters. These data are deeply fused using an extended Kalman filter to calculate the precise position of the UAV's center of mass in the ENU coordinate system. and its rotation matrix from the slave system to the ENU system .

[0128] Based on this, the cleaning robot's position in the ENU coordinate system Calculated using the following formula:

[0129]

[0130] in These are the coordinates of the robot's marker origin in the body coordinate system, derived from... The extracted data is obtained. This formula completes the expression of the vector from the origin of the organism to the origin of the marker in the ENU system.

[0131] For solving the robot's global orientation, its rotation matrix Determined by the following formula:

[0132]

[0133] S4: Based on the multimodal fuzzy PID control method, the flight speed of the UAV is dynamically adjusted according to the real-time pose of the cleaning robot, so as to achieve stable tracking and position maintenance of the UAV on the cleaning robot.

[0134] Specifically, the steps include:

[0135] S41: Position Error Calculation and Control Mode Switching

[0136] Visual positioning data reception and processing: The onboard computer of the drone receives the relative position of the cleaning robot in the body coordinate system in real time from the visual positioning system. This position information is represented in the form of a three-dimensional vector.

[0137]

[0138] in, These represent the front-back, left-right, and up-down position components of the cleaning robot relative to the drone's coordinate system. The visual positioning system calculates the target position in real time using image processing algorithms, providing data support for subsequent control decisions.

[0139] Position error calculation: The system calculates the current position error and compares the actual detected target position with the expected tracking hover point.

[0140]

[0141] The desired tracking hover point is defined as:

[0142]

[0143] in, The preset hovering height ensures the drone maintains an appropriate distance directly above the cleaning robot. Position error vector. This reflects the deviation between the current tracking state and the ideal state.

[0144] The system extracts the horizontal position error and calculates its norm:

[0145]

[0146] By comparing the horizontal error norm with a preset threshold, the system can autonomously determine and switch control modes. This dual-mode control strategy uses the optimal control method under different distance conditions.

[0147] Remote position loop PID control: when When the threshold is reached, the system activates position loop PID control, employing a cascaded PID structure to quickly reduce the distance to the target:

[0148]

[0149] in, For speed control commands, For positional error, These are the proportional, integral, and derivative gains of the position loop. This control mode prioritizes response speed, helping the UAV quickly approach the target.

[0150] Close-range speed loop PID control: when When the threshold is reached, the system switches to speed loop PID control for precise positioning via feedforward.

[0151]

[0152] in, For acceleration control commands, For the desired speed, For actual speed, These are the speed loop control parameters. This mode ensures that the drone can accurately track the movement of the cleaning robot.

[0153] Expected speed Based on the pose changes of the cleaning robot, dynamic prediction is performed, and the pose sequence of the cleaning robot in the body coordinate system is obtained in step S3. A filter is used to estimate the instantaneous velocity of the cleaning robot relative to the drone.

[0154] The filter design is as follows:

[0155]

[0156] in, It is the predicted position at time k. This is the filter gain. Ultimately, the filter output... This is the estimated velocity of the target relative to the aircraft. It represents the expected velocity of the UAV in the aircraft's coordinate system. It should be set to have the same magnitude and opposite direction as the target's predicted velocity in order to counteract its motion.

[0157]

[0158] S42: Fuzzy Controller Construction

[0159] The first input variable of the fuzzy logic system is the normalized position error, which is calculated using the following formula:

[0160]

[0161] in, for The horizontal position error norm at time t. The maximum permissible error is used for normalization. This is the scaling factor. This normalization process ensures that the system maintains consistency and robustness under different flight conditions.

[0162] The second input variable is the normalized error rate of change, which reflects the dynamic trend of error change:

[0163]

[0164] in, Sampling time, This is the scaling factor. The error change rate predicts the future trend of the system, providing forward-looking information for adjusting control parameters.

[0165] Fuzzy subsets and membership functions:

[0166] for and Define the same fuzzy subset: negative large small burden ,zero , Zheng Xiao Zhengda

[0167] A Gaussian membership function is used to transform precise input values ​​into membership degrees for various fuzzy languages:

[0168]

[0169] in, The center of the Gaussian function, This is the standard deviation parameter.

[0170] Fuzzy rule base design: based on and The fuzzy subsets are combined to establish an IF-THEN fuzzy rule base, with the consequent being the PID parameter increment. , A fuzzy subset of output.

[0171] proportional gain Taking the tuning rules as an example:

[0172] when for IF is THEN is ;

[0173] when for and for IF is AND is THEN is ;

[0174] when for and for IF is AND is THEN is , etc., for Design 25 lines respectively Rules, forming a complete rule table For example ( Figure 5 ).

[0175] Mamdani-type inference and defuzzification: Employing a Mamdani-type inference engine, the activation strength of each rule... The minimum membership degree of the predecessor:

[0176]

[0177] Aggregate the outputs of all activated rules and calculate the centroid of the aggregated fuzzy set using the centroid method:

[0178]

[0179] in, represent Any increment in, This is for quantifying the number of levels.

[0180] In each control cycle The parameters of the PID controller are updated online using the following formula:

[0181]

[0182] This adaptive adjustment mechanism enables the control system to optimize its control performance based on real-time conditions.

[0183] S43: Control Mode Selection and Command Generation

[0184] The UAV flight control system outputs corresponding control commands based on the control mode selected in step S41. The system intelligently switches between position control mode and speed control mode according to the current tracking status and error magnitude.

[0185] The position error calculated in step S41 is directly converted into a velocity command:

[0186]

[0187] This speed command drives the drone to move towards the target point, enabling rapid approach over a wide area. The position control mode is suitable for situations where the drone and the cleaning robot are far apart, and can quickly reduce tracking errors.

[0188] Speed ​​control mode command processing: In speed control mode, the system aims to maintain zero relative speed and calculates the speed error.

[0189]

[0190] The system maps speed errors to pitch, roll, and throttle commands for the UAV, enabling precise relative position maintenance.

[0191] Dynamics decoupling and command mapping: Control commands undergo dynamics decoupling processing, corresponding to different degrees of freedom of motion of the UAV.

[0192]

[0193] The pitch command controls the drone's forward and backward movement, the roll command controls its left and right movement, and the throttle command controls its vertical movement. This decoupled design ensures the independence and precision of control for each axis.

[0194] Closed-loop control and real-time adjustment: The entire command transmission process forms a complete closed-loop control system. The flight control system continuously adjusts the control quantities of each channel at a millisecond frequency, applying the PID parameters optimized in step S42 to the control calculations in real time.

[0195] System stability assurance: Through the adaptive adjustment of parameters in the fuzzy logic system and the coordinated operation of the dual-modal control strategy, the system ensures that the UAV maintains both agile response and flight stability during tracking. The control system can adapt to different flight environments and tracking conditions, achieving robust autonomous tracking performance.

[0196] S5: Through cross-domain collaborative control between the onboard computer and the mechanical gripper system, the mechanical gripper is driven to perform grasping operations, and the grasping status is determined by the current signal fed back by the electromagnet adsorption module.

[0197] Specifically, the steps include:

[0198] S51: Grasp Position Determination: After the UAV enters the target area, the onboard computer continuously monitors the relative pose data from the visual positioning system. The system sets a grasp position determination threshold, whereby the horizontal positioning error must be less than 5 cm and the attitude angle deviation must be controlled within 3 degrees. To eliminate transient jitter interference, the system requires that the above pose data must continuously and stably meet the threshold conditions for a duration of 2 seconds. Simultaneously, the system calculates the dynamic characteristics of the UAV through the IMU unit, requiring the angular velocity vector magnitude to be less than 0.1 radians / second and the linear acceleration fluctuation range to be controlled within ±0.2g. When both the spatial pose conditions and the dynamic stability conditions are met, the system generates a descent authorization command and initiates the grasp process.

[0199] S52: Height Descent and Safety Monitoring under Cascaded Control:

[0200] After obtaining capture authorization, the drone initiates a cascaded control descent procedure. The decision-making layer generates a smooth vertical descent velocity curve based on real-time distance information provided by the visual positioning system. This curve employs a trapezoidal velocity programming algorithm, setting an acceleration limit of 0.1 m / s² to ensure a smooth and shock-free descent. The execution layer tracks this velocity command through a lower-level altitude PID controller with a control cycle of 10 ms, achieving millimeter-level altitude control accuracy.

[0201] The safety monitoring system employs a priority interruption mechanism. The horizontal position monitoring module continuously calculates the XY plane error, triggering a protection mechanism when the error in any direction exceeds 8 cm. The attitude stability monitoring module calculates the body angular velocity using IMU data; if the roll or pitch angular velocity exceeds 0.15 rad / s, it is determined to be instability. If the above mechanism is triggered, the flight control system executes an emergency ascent strategy, returning to the initial hovering altitude at a vertical speed of 0.3 m / s.

[0202] S53: Electromagnetic induction triggering and grasping execution:

[0203] The drone descends to a point 10 centimeters above the target location, and its onboard system pre-activates a miniature electromagnet. This compact electromagnet operates at 5V and generates a guiding magnetic field of 0.15T at close range. As the robotic gripper gradually approaches the magnetic material on top of the cleaning robot, the magnetic field generates a lateral corrective force of approximately 0.5N during the final stage of its movement. This micro-force is sufficient to compensate for any final alignment deviation, achieving passive compliant positioning.

[0204] When the electromagnet comes into contact with the magnetic conductor, the change in magnetic reluctance causes the coil current to rise from 80mA to 120mA. A high-precision current detection circuit detects this 40mA jump, and when the signal exceeds the 100mA threshold for 10ms, the system confirms successful positioning. The onboard computer then triggers a grasping command, driving a micro-servo motor to close the mechanical gripper. The electromagnet is immediately de-energized after grasping; its function is limited to positioning guidance and status feedback, and it does not participate in subsequent fixed support.

[0205] S6: Conduct statistical analysis on the success rate of recovery, positioning accuracy, grasping stability and safety to verify the reliability and operational efficiency of the autonomous recovery method.

[0206] Specifically, the steps include:

[0207] Twenty repetitive experiments were conducted to test the success rate of the overall process and verify the reliability of the invention. During the experiments, three failures occurred: one due to strong light interference, one due to a sudden gust of wind, and the other due to excessive speed caused by the PID control mode not switching in time. In all three cases, the system safely aborted the operation and initiated the recovery process. These results preliminarily verify the basic reliability of the system under limited experimental conditions.

[0208] RTK positioning accuracy analysis was performed using the elliptic error method. By measuring the RTK latitude and longitude coordinates of the same location, the directional systematic error of the RTK positioning system was calculated. The calculated average positioning error was 3.98 cm. The elliptic error analysis is as follows: Figure 7 .

[0209] The positioning accuracy and tracking stability of the UAV using ArUco codes were analyzed. The UAV positioning response curve and corresponding data are shown below. Figure 8 The response curve and corresponding data of drone tracking are as follows: Figure 9 It can be seen that its positioning accuracy is high, its steady-state error is small, and its response speed is fast, but its overshoot is high and needs further improvement.

[0210] Crawling stability analysis shows that, Figure 10 The yaw angle of the payload exhibits a relatively large rate of change within its Euler angles, while the roll and pitch angles remain within a small range. This indicates that the robotic gripper effectively suppresses the load's swaying within the flight plane, providing excellent lateral and longitudinal stability. From a safety perspective, the screw drive mechanism possesses a self-locking function, fundamentally preventing load slippage. Mechanical testing has verified that its transmission system can withstand an axial force of no less than 400N.

[0211] The above embodiments are merely typical illustrative methods of the present invention, and the scope of protection of the present invention is not limited thereto. All equivalent substitutions and improvements made under the concept of the present invention should fall within the scope of protection. It should be emphasized that any modifications or minor adjustments made by those skilled in the art without departing from the basic principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A control method for a drone-based autonomous photovoltaic module cleaning robot, characterized in that, Includes the following steps: S1: The ground base station sends the target photovoltaic panel location and task information to the drone through a task planning algorithm. The drone flies to the target photovoltaic panel and hovers at the designated height through position loop control. S2: Before or periodically perform camera calibration on the airborne vision system to obtain the camera's intrinsic parameter matrix and distortion coefficients for error correction in subsequent ArUco image recognition and position calculation. S3: Activate the onboard vision system, identify the ArUco code carried on the top of the cleaning robot, calculate its relative position, rotation matrix and robot number in the image coordinate system and the body coordinate system, and convert the pose information of the cleaning robot to the ENU coordinate system through coordinate transformation to obtain the precise position in the global coordinate system. S4: Based on the multimodal fuzzy PID control method, the flight speed of the UAV is dynamically adjusted according to the real-time pose of the cleaning robot, so as to achieve stable tracking and position maintenance of the UAV on the cleaning robot. S5: Through cross-domain collaborative control between the onboard computer and the mechanical gripper system, the mechanical gripper is driven to perform grasping operations, and the grasping status is determined by the current signal fed back by the electromagnet adsorption module. S6: Analyze the success rate of recovery, positioning accuracy, grasping stability and safety to verify the reliability and operational efficiency of the autonomous recovery method.

2. The control method for a drone-based autonomous photovoltaic module cleaning robot according to claim 1, characterized in that, Step S1 includes: S1.1: Task Information Perception and Reporting: The inspection drone inspects the photovoltaic array through its onboard vision system, identifies and records the numbers and GNSS coordinates of dirty and clean photovoltaic modules; at the same time, it detects whether there are cleaning robots working on the clean photovoltaic modules; subsequently, the inspection drone uploads data packets containing "location information of dirty photovoltaic modules that need cleaning" and "location information of clean photovoltaic modules with idle cleaning robots" to the ground base station via wireless data link; S1.2: Mission Planning and Command Issuance: After receiving the inspection data, the ground base station performs mission scheduling and planning, calculates the optimal recovery and deployment path, generates flight mission instructions for the transport drone, which explicitly include the "target recovery location" and the "target deployment location"; the base station issues the mission instructions to the designated transport drone through the wireless communication link. S1.3: Autonomous Flight and Hovering of the UAV: ​​The transport UAV receives and parses the instructions from the base station, uses its built-in GNSS receiver and inertial measurement unit (IMU) for position and attitude feedback, and adopts a PID position loop control algorithm to generate position and velocity control instructions in three-dimensional space. The UAV first flies to the "target recovery position", and through this closed-loop control, it stably flies to the top of the target photovoltaic module and achieves precise hovering at the predetermined height, preparing for subsequent visual recognition and grasping operations.

3. The control method for a drone-based autonomous photovoltaic module cleaning robot according to claim 1, characterized in that, Step S2 prepares for subsequent ArUco image recognition and location calculation: S2.1: Camera intrinsic parameter calibration: Using a checkerboard calibration board, the camera's intrinsic parameter matrix and lens distortion coefficients are calculated and obtained from images acquired from multiple angles, thus completing the internal parameter calibration of the vision system; S2.2: Establishment of the extrinsic coordinate system: Establish the ENU coordinate system with the UAV takeoff point as the origin, and solve the extrinsic parameters of the camera relative to this coordinate system through spatial registration, including the rotation matrix R and the translation vector T, to determine the attitude and position of the camera in global space; S2.3: Shooting posture planning and adjustment: Based on the actual installation tilt angle of the photovoltaic modules, plan the flight altitude and trajectory of the drone, and control the gimbal mechanism to make the camera optical axis approximately perpendicular to the ground.

4. The control method for a drone-based autonomous photovoltaic module cleaning robot according to claim 1, characterized in that, Step S3 includes: S3.1: ArUco code detection and information extraction: Activate the onboard vision system to process the real-time acquired images; identify the ArUco code carried on the top of the cleaning robot, extract its corner pixel coordinates and read its unique number; S3.2: Solve the relative pose in the camera coordinate system: Using the intrinsic parameter matrix and distortion coefficients obtained in advance through camera calibration, combined with the known actual physical size of the ArUco code, the three-dimensional relative position and rotation matrix of the ArUco code in the camera coordinate system are solved by the PnP algorithm, so as to obtain the precise pose of the robot relative to the camera. S3.3: Coordinate transformation to body coordinate system: Based on the fixed installation relationship between the camera and the UAV body, the relative pose in the camera coordinate system calculated by S3.2 is transformed to the UAV body coordinate system using the camera-body extrinsic parameters; S3.4: Transform to the global ENU coordinate system: By integrating the UAV's own GNSS position and IMU attitude data, the relative pose obtained in the body coordinate system in S3.3 is further transformed to the ENU global coordinate system with the takeoff point as the origin, and finally the absolute position and global orientation of the cleaning robot are obtained, thus completing its precise positioning in the world coordinate system.

5. The control method for a drone-based autonomous photovoltaic module cleaning robot according to claim 1, characterized in that, Step S4 includes: S4.1: Modal determination: The UAV's onboard computer receives the relative position of the cleaning robot in the body coordinate system calculated by the onboard vision system in real time. ; Calculate the position error at the current moment ,in The desired hovering point for the drone; based on the norm of the horizontal error. The system compares the data with a preset threshold, automatically determines and switches the control mode, using position loop PID control for long distances and speed loop PID control for short distances. S4.2: Fuzzy PID Parameter Tuning: To overcome the shortcomings of traditional PID parameters being fixed and having poor adaptability in complex environments, a fuzzy logic controller is introduced for each control mode. This controller takes the normalized position error E and error change rate ΔE as inputs, performs inference through a predefined fuzzy rule set, and outputs the PID control parameters online in real time. The adjustment amount; this mechanism enables the control system to dynamically adjust its control strategy according to different tracking states, combining responsiveness and stability; S4.3: Multimodal control law calculation and execution: Based on the currently active control mode and the fuzzy tuned PID parameters, the final control command is calculated. The generated control command is sent to the UAV's flight control system to drive the UAV to achieve dynamic tracking, thus completing stable tracking and position holding.

6. The control method for a drone-based autonomous photovoltaic module cleaning robot according to claim 1, characterized in that, Step S5 includes: S5.1: Grasping Position Determination: The onboard computer continuously monitors the relative pose of the UAV-robot from S4; it sets the grasping position threshold based on the horizontal error and attitude angle, and authorizes the entry into the descent grasping phase when the conditions are met; Spatial conditions: The current relative pose error remains stable within the grasping position threshold for more than 2 seconds; Dynamic conditions: The sway amplitude of the UAV calculated from the IMU data is lower than the stability threshold. S5.2: Altitude Descent and Safety Monitoring under Cascaded Control: The decision-making layer generates a smooth vertical descent speed command based on the visual positioning results; the execution layer, controlled by the UAV flight controller, precisely executes the altitude descent through the underlying PID controller; simultaneously, the outer-loop decision-making layer continuously monitors safety: real-time detection of horizontal position error and attitude stability; if the error exceeds the safety threshold or the stability is insufficient, the descent is immediately interrupted, triggering an emergency ascent command to return the UAV to a safe altitude and re-perform the grab-and-position judgment, thereby ensuring operational safety under external interference; S5.3: Electromagnetic induction triggering and gripping execution: When the drone descends close to the cleaning robot, the onboard computer pre-activates the electromagnet at the bottom of the mechanical claw to generate a magnetic field; the magnetic field forms a lateral correction force when it approaches the metal object to guide the mechanical claw to accurately align and achieve passive compliant positioning; once the electromagnet attracts the object, the circuit current rises significantly, and the onboard computer determines that the attraction is successful and then instructs the mechanical claw to close, ensuring reliable transport connection.

7. The control method for a drone-based autonomous photovoltaic module cleaning robot according to claim 1, characterized in that, Step S6 includes: Recovery success rate: The statistics are based on a large number of repetitive trials, recording the frequency of successful grabbing and improved transfer, and comparing it with the total number of attempts to calculate the baseline recovery success rate; Absolute and relative accuracy of positioning: Absolute accuracy is determined by comparing the deviation between the robot's global coordinates calculated by vision and the high-precision differential GNSS measurement reference value; relative accuracy is determined by analyzing the fluctuation range and convergence characteristics of the relative pose data of the UAV and the cleaning robot during the hovering phase. Grasping stability and safety: The Euler angles of the transported object are measured by the IMU measurement module to determine the stability of the gripping; the maximum axial force of the mechanical gripper is measured to determine the upper limit of the weight it can bear and to evaluate the safety of the gripping process.

Images