Crawler-type cleaning robot path planning and blind area elimination method

Abstract

The present application relates to the technical field of automatic control and machine vision, and particularly relates to a path planning and blind area elimination method for a crawler-type cleaning robot, which comprises the following steps: collecting a photovoltaic module surface image through a vision sensor arranged on a robot body, extracting photovoltaic cell grid line features and establishing a local coordinate system; generating a path reference line based on the grid line direction, and calculating a lateral deviation and an angle deviation between a current position of the robot and the path reference line in real time; dynamically adjusting a robot walking path through a fuzzy control method; dynamically adjusting a cleaning interval by calculating a current cleaning width and an expected overlap rate; and executing an obstacle crossing or board crossing decision based on an inclination angle data when sensor data conflicts. A multi-sensor data fusion technology improves the anti-interference capability of the robot, so that the robot can stably operate in a strong light, a reflective light and the like.

Description

Technical Field

[0001] This invention relates to the field of interdisciplinary technology of automatic control and machine vision, and in particular to a method for path planning and blind spot elimination of a tracked cleaning robot. Background Technology

[0002] As a crucial component of clean energy, photovoltaic (PV) power generation efficiency is directly affected by the cleanliness of the module surface. PV modules are exposed to the natural environment for extended periods, easily accumulating dust, bird droppings, snow, and other contaminants on their surfaces. These stains significantly reduce photoelectric conversion efficiency. Studies have shown that severe dirt accumulation can decrease power generation efficiency by more than 25%. Therefore, regular cleaning and maintenance of PV modules is a key aspect of ensuring the power generation efficiency of a power plant.

[0003] Currently, tracked cleaning robots are gradually becoming an important tool for cleaning and maintaining photovoltaic power plants due to their excellent obstacle-crossing ability and terrain adaptability. However, existing tracked cleaning robots still face several technical bottlenecks in practical applications: First, in terms of path planning, since photovoltaic arrays are often irregularly arranged due to terrain undulations, installation errors, or differences in component size, traditional path planning methods based on pre-set tracks or simple edge recognition are difficult to achieve cross-array compatibility. This necessitates specialized debugging of the robot for different arrays, significantly reducing its efficiency. Second, regarding cleaning effectiveness, robots are prone to missing areas when turning, crossing panels, or at edges, especially when there are height differences or gaps between components, where blind spots are particularly prominent. Furthermore, in terms of dynamic obstacle avoidance, existing sensor solutions (such as infrared or ultrasonic sensors) are prone to misjudgment under strong light, reflections, or dirt interference, leading to robot positioning deviations or even falls, seriously affecting operational safety and reliability.

[0004] These technical problems severely restrict the widespread application of tracked cleaning robots. This is especially true in large-scale photovoltaic power plants, where the large number of arrays, wide distribution, and complex environment exacerbate these issues. Therefore, developing a path planning and blind spot elimination method that can adapt to irregular arrays, effectively eliminate cleaning blind spots, and possess strong anti-interference capabilities has become a pressing technical challenge in this field. Summary of the Invention

[0005] To achieve the above objectives, the present invention provides a method for path planning and blind spot elimination for a tracked cleaning robot, comprising the following steps:

[0006] Images of the photovoltaic module surface are acquired by a vision sensor installed on the robot body, the grid line features of the photovoltaic cells are extracted, and a local coordinate system is established.

[0007] A path baseline is generated based on the grid line direction, and the lateral and angular deviations between the robot's current position and the path baseline are calculated in real time.

[0008] Based on the lateral and angular deviations, the robot's walking path is dynamically adjusted using fuzzy control.

[0009] A map of the cleaned area is generated based on historical cleaning trajectory records, and the cleaning spacing is dynamically adjusted by calculating the current cleaning width and the expected overlap rate.

[0010] By fusing data from multiple sensors to identify obstacles, when sensor data conflict, obstacle crossing or board crossing decisions are made based on tilt angle data.

[0011] Preferably, the specific steps for extracting the grid line features of the photovoltaic cell and establishing a local coordinate system include:

[0012] The color images acquired by the vision sensor are converted to grayscale, transforming the RGB color space into a grayscale space.

[0013] An adaptive threshold segmentation method is used to binarize grayscale images. The threshold of the adaptive threshold segmentation method is dynamically determined based on the grayscale distribution characteristics of local regions of the image.

[0014] The grid lines of photovoltaic cells are identified by detecting straight line features in a binary image using Hough transform.

[0015] The least squares method is used to fit the equations of the detected grid lines, and the coordinates of the intersection points between the grid lines are calculated.

[0016] Choose the intersection of the grid lines closest to the image center as the origin of the local coordinate system, and use the direction of grid line extension as the axis of the coordinate system.

[0017] Preferably, the specific method for calculating the lateral and angular deviations between the robot's current position and the path baseline in real time includes:

[0018] The robot's current position image is acquired in real time using a vision sensor, and the offset of the current position relative to the grid lines is identified.

[0019] The vertical distance from the robot's center point to the path baseline is calculated as the lateral deviation. The path baseline is a reference line that is parallel to the grid lines and passes through the center of the ideal path.

[0020] The angle between the robot's current heading angle and the path baseline direction is calculated as the angle deviation. The heading angle is obtained through attitude data provided by the inertial measurement unit.

[0021] The values ​​of lateral deviation and angular deviation are continuously updated at a fixed sampling period to form deviation trend data.

[0022] Preferably, the specific implementation method of dynamically adjusting the robot's walking path through fuzzy control includes:

[0023] A fuzzy controller is established with lateral deviation and angular deviation as input variables and track speed compensation value as output variable;

[0024] The fuzzy subsets of the input variables are defined to include five levels: negative large, negative small, zero, positive small, and positive large. The membership function of each fuzzy subset adopts a triangular function.

[0025] A fuzzy control rule base is established based on experimental calibration. The fuzzy control rule base includes a track speed compensation strategy based on the combination of lateral deviation and angular deviation.

[0026] The accurate left and right track speed compensation values ​​are obtained by defuzzification calculation using the center of gravity method, and then the left and right track speed compensation values ​​are superimposed on the basic drive speed.

[0027] The output gain coefficient of the fuzzy controller is dynamically adjusted based on the deviation change trend data.

[0028] Preferably, the specific method for dynamically adjusting the cleaning spacing by calculating the current cleaning width and the expected overlap rate includes:

[0029] The effective cleaning width is calculated based on the actual ground contact length of the brush track and the robot's current heading angle, which is measured in real time by an inertial measurement unit;

[0030] By analyzing historical cleaning trajectory data, the distribution of cleaned and uncovered areas can be statistically analyzed.

[0031] The minimum overlap threshold required is calculated based on the ratio of the uncovered area to the current cleaning width;

[0032] The overlap rate threshold is dynamically adjusted based on the degree of surface contamination of the photovoltaic module. The degree of contamination is determined by the image grayscale variance analyzed by a visual sensor.

[0033] The overlap rate threshold calculated in real time is used as the basis for adjusting the spacing of the next cleaning cycle.

[0034] Preferably, the dynamic adjustment method for the overlap rate threshold further includes:

[0035] When the area of ​​the uncovered region is detected to be continuously increasing, the overlap rate threshold is increased in a stepwise manner.

[0036] When no uncovered areas are found in multiple consecutive cleaning cycles, the overlap rate threshold is gradually reduced.

[0037] Set upper and lower limits for the overlap rate threshold, which are determined based on the size of the photovoltaic module and the robot's structural parameters;

[0038] The sensitivity of the overlap rate threshold is adjusted based on the alignment of the photovoltaic array, and the alignment is calculated by the grid line fitting error.

[0039] Preferably, the specific steps for obstacle identification by fusing multi-sensor data include:

[0040] The edge features of the photovoltaic panel are identified by a visual sensor, and the geometric parameters of the edge contour are extracted.

[0041] The distance information of obstacles in front of the robot is detected by an infrared ranging sensor, and the distance information includes a minimum distance value and an average distance value.

[0042] The robot's tilt angle is monitored in real time using an inertial measurement unit, and the rate of tilt angle change is recorded.

[0043] A multi-sensor data fusion decision model is established, which first evaluates the confidence level of each sensor's data.

[0044] When there are conflicting results from different sensors, the tilt angle data provided by the inertial measurement unit shall be given priority as the basis for decision-making.

[0045] Preferably, the specific method for making obstacle-crossing or board-crossing decisions based on tilt angle data includes:

[0046] When the tilt angle is detected to be continuously increasing and exceeding the safety threshold, the obstacle crossing mode is triggered. The safety threshold is determined based on the maximum allowable tilt angle of the photovoltaic module.

[0047] In obstacle-crossing mode, the control system increases the output torque of the track drive motor while reducing the travel speed;

[0048] When a sudden change in tilt angle is detected and the distance value displayed by the infrared sensor increases sharply, the cross-board mode is triggered;

[0049] In cross-plate mode, the cleaning function is paused and the body posture is adjusted. Stable cross-plate movement is achieved by controlling the differential speed of the two side tracks.

[0050] After completing the obstacle crossing or board crossing action, it automatically resumes the normal cleaning operation mode.

[0051] Preferably, after dynamically adjusting the cleaning spacing, a blind spot compensation step is also included:

[0052] Additional cleaning paths are generated based on the distribution characteristics of uncovered areas in historical cleaning trajectories.

[0053] The additional cleaning path adopts a reciprocating motion mode, and the motion range is determined according to the area and shape of the uncovered area;

[0054] During the execution of the additional cleaning path, the cleaning coverage effect is monitored in real time until the predetermined cleanliness standard is reached;

[0055] Record the parameters of this additional cleaning path as a reference for path planning in similar situations in the future.

[0056] Preferably, a data reporting step is included after the obstacle-crossing or board-crossing decision is made:

[0057] The cleaning operation data is uploaded to the monitoring center via a wireless communication module;

[0058] The cleaning operation data includes the actual cleaning coverage rate, a map of uncovered areas, and an obstacle handling log.

[0059] The actual cleaning coverage rate is calculated as the ratio of the cleaned area to the total surface area of ​​the photovoltaic modules.

[0060] Generate a cleaning operation report, including an effectiveness assessment of the operation and suggestions for improvement.

[0061] The beneficial effects of this invention are:

[0062] 1. This invention acquires images of the photovoltaic module surface using a vision sensor, extracts the grid line features of the photovoltaic cells, establishes a local coordinate system, and generates a path baseline based on the grid line direction. The lateral and angular deviations of the robot's current position from the path baseline are calculated in real time, and the walking path is dynamically adjusted. This method can automatically adapt to terrain changes in different photovoltaic arrays, eliminate the limitations of traditional path planning methods, achieve cross-array versatility, and improve robot efficiency and adaptability.

[0063] 2. This invention dynamically adjusts the cleaning spacing, generates a map of the cleaned area by combining historical cleaning trajectory data, calculates the current cleaning width and the expected overlap rate, and adjusts the cleaning spacing in real time, effectively avoiding missed areas. Furthermore, the generation of additional cleaning paths and blind spot compensation steps ensure the resolution of blind spot problems, accurately covering all uncleaned areas and greatly improving cleaning efficiency and effectiveness.

[0064] 3. This invention employs multi-sensor data fusion technology, integrating data from visual sensors, infrared sensors, and inertial measurement units, and combining this with tilt angle data to make obstacle-crossing or board-crossing decisions. Through confidence assessment and a decision-making mechanism that prioritizes tilt angle data, sensor misjudgment is effectively avoided, improving the robot's anti-interference capability in complex environments. This technology significantly enhances the robot's operational safety and reliability, ensuring stable operation even in environments with strong light, reflections, or dirt interference. Attached Figure Description

[0065] To more clearly illustrate the technical solutions in this invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0066] Figure 1 This is a flowchart of the steps of the method of the present invention;

[0067] Figure 2 This is a flowchart of the blind zone compensation step in the method of the present invention;

[0068] Figure 3 This is a flowchart of the data reporting steps in the method of the present invention;

[0069] Figure 4 This is a schematic diagram of the tracked cleaning robot of the present invention. Figure 1 ;

[0070] Figure 5 This is a schematic diagram of the tracked cleaning robot of the present invention. Figure 2 ;

[0071] Figure 6 This is a schematic diagram of the tracked cleaning robot of the present invention. Figure 3 . Detailed Implementation

[0072] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. It should also be noted that, to make the embodiments more comprehensive, the following embodiments are the best and preferred embodiments, and those skilled in the art can use other alternative methods to implement some well-known technologies; moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0073] Please see Figures 1-6 This invention provides a method for path planning and blind spot elimination for a tracked cleaning robot. First, a high-resolution vision sensor is installed on the tracked cleaning robot to capture images of the photovoltaic module surface in real time. Image processing algorithms extract the grid line features on the photovoltaic cells, and image recognition technology accurately identifies the direction of the grid lines. Based on these features, a local coordinate system is established. This step provides a precise geographic reference frame for subsequent path planning, ensuring that the robot can clean in the correct direction.

[0074] It is understood that the tracked cleaning robot of the present invention includes at least the following systems:

[0075] 1. Walking system:

[0076] Tracks: High-strength, wear-resistant rubber or composite material tracks. The track surface is designed with a special pattern to enhance adhesion to the glass panel and prevent slippage.

[0077] Drive motors: Two high-torque DC servo motors drive the tracks on both sides respectively, and the differential steering enables flexible forward, backward, turning and rotation in place.

[0078] Suspension system: Designed with a simple independent suspension mechanism to ensure that all track segments remain in contact with the surface when the robot is on uneven surfaces (such as slightly deformed surfaces) or crossing edges, providing continuous driving force.

[0079] 2. Cleaning system:

[0080] Brush Track: Located at the bottom of the robot, this motor-driven brush track features soft nylon or PP bristles to prevent scratching the glass. The brush track combines the cleaning power of a roller brush with the wide coverage area of ​​a track, allowing for cleaning a larger area in a single pass.

[0081] Spraying device: Includes a miniature water pump and atomizing nozzles. The miniature water pump pressurizes the water or cleaning solution, which is then sprayed onto the photovoltaic panel glass surface through the atomizing nozzles. This greatly reduces water consumption while ensuring cleaning effectiveness.

[0082] 3. Control system:

[0083] Main control module (MCU): such as STM32 series or higher-level processors, serving as the "brain" of the robot.

[0084] Navigation and positioning: An integrated vision sensor (camera) is used to correct the path by recognizing the grid strips of the photovoltaic panel cells.

[0085] Perception and obstacle avoidance: Infrared sensors are distributed around the body to detect the edges of photovoltaic panels and prevent them from falling.

[0086] 4. Energy System:

[0087] Solar charging panels: Solar panels are installed on the top of the robot, which can continuously replenish its power while it is working.

[0088] High-capacity lithium batteries: used as the main power source.

[0089] 5. Security and Communication Systems:

[0090] Wireless communication module: 4G and LoRa modules are used to communicate with the remote monitoring center, upload information such as working status, location, power level, and fault codes, and receive control commands.

[0091] Emergency Stop Button: A physical button to ensure that maintenance personnel can manually stop the system in an emergency.

[0092] Based on the extracted grid line features, a path baseline is generated and used as a reference line for the robot's movement. The robot monitors its current position in real time using sensors and compares it with the path baseline, calculating the lateral and angular deviations between the current position and the baseline. This process accurately tracks the robot's trajectory, ensuring that the robot always moves along the predetermined path.

[0093] Based on real-time calculated lateral and angular deviations, a fuzzy control algorithm is used to dynamically adjust the robot's walking path. Fuzzy control can adjust the robot's direction of travel according to different degrees of deviation, enabling the robot to smoothly clean along irregular array surfaces while avoiding path deviation or off-tracking, thus improving the accuracy and flexibility of path planning.

[0094] During the cleaning process, the robot generates a map of the cleaned areas based on historical cleaning trajectory records. The system dynamically adjusts the cleaning spacing according to the current cleaning width and the expected overlap rate. If it finds that the overlap rate of some areas is insufficient or that the cleaning is not completely covered, the robot will automatically adjust the spacing of the cleaning path to ensure that no cleaning area is missed, thereby effectively avoiding blind spots and improving cleaning results.

[0095] The robot is equipped with multiple sensors (such as vision sensors, infrared sensors, and ultrasonic sensors) to perceive surrounding obstacles in real time. Through sensor data fusion technology, when conflicting sensor readings occur, the system prioritizes tilt angle data to determine if the robot has encountered an obstacle and makes obstacle-crossing or board-crossing decisions. This mechanism ensures the robot can safely traverse obstacles in complex environments, especially in unevenly spaced arrays, preventing falls and improving operational safety and reliability.

[0096] This invention effectively solves problems such as irregular photovoltaic arrays, blind spots in cleaning, and dynamic obstacle avoidance by combining visual sensors with fuzzy control technology. The robot can automatically adapt to different photovoltaic array configurations, ensuring thorough cleaning coverage and stable operation in complex terrain. Simultaneously, multi-sensor data fusion technology improves the robot's anti-interference capabilities, enabling stable operation in environments with strong light and reflections. Through these innovative technologies, this invention significantly improves cleaning efficiency and operational safety, providing a more intelligent and efficient solution for the cleaning and maintenance of photovoltaic power plants.

[0097] In one possible implementation, the color image (typically an RGB image) acquired by the vision sensor is first converted to a grayscale image. Grayscale conversion eliminates interference from color information while preserving the image's brightness information, thus simplifying subsequent image processing and reducing computational complexity. Each pixel value in the grayscale image represents brightness information, facilitating subsequent feature extraction.

[0098] An adaptive thresholding method is used to binarize grayscale images, separating the photovoltaic cell grid lines from the background region. Traditional fixed thresholding methods may be affected by environmental factors such as changes in lighting, leading to inaccurate segmentation. In contrast, the adaptive thresholding method dynamically adjusts the threshold based on the grayscale distribution characteristics of local image regions, thus stably segmenting grid line features under different lighting conditions and improving the robustness of the algorithm.

[0099] Next, the Hough transform is used to detect straight lines in the binarized image. The Hough transform is a classic image processing algorithm capable of detecting straight line features in complex backgrounds. This step accurately identifies the grid lines of the photovoltaic cells and provides crucial data for subsequent coordinate system establishment.

[0100] The linear features detected by the Hough transform are fitted using the least squares method to calculate the accurate equations of the grid lines. The least squares method is a commonly used fitting technique that minimizes the fitting error, ensuring that the calculated lines more accurately reflect the grid line positions of the actual photovoltaic module. This step effectively improves the accuracy of grid line extraction, providing more reliable basic data for path planning.

[0101] Based on the fitted grid line equations, the coordinates of the intersection points between the grid lines are calculated. Then, the intersection point closest to the image center is selected as the origin of the local coordinate system. This selection method ensures that the coordinate system is established at the center of the image, reducing bias. Furthermore, the extension direction of the grid lines will serve as the axis of the coordinate system, ensuring accurate tracking of the orientation and position of the photovoltaic cells during path planning.

[0102] This invention effectively extracts the grid line features of photovoltaic cells and accurately establishes a local coordinate system. Compared with traditional methods, the adaptive threshold segmentation method and Hough transform can better handle complex lighting and environmental changes, ensuring the accuracy of image processing. The application of the least squares method improves the accuracy of grid line extraction, while selecting the intersection point closest to the image center as the origin of the coordinate system further improves the stability of path planning. Through these techniques, this invention can provide accurate coordinate references for subsequent path planning, reduce deviations and blind spots during the cleaning process, and greatly improve cleaning efficiency and operational safety.

[0103] In one possible implementation, the robot is equipped with a vision sensor (such as a camera) to capture image information of its current position in real time. Image processing algorithms analyze and identify the positions of the photovoltaic cell grid lines in the image, and further calculate the robot's current position offset relative to the grid lines. This process accurately determines whether the robot has deviated from the ideal path, providing raw data for subsequent deviation calculations.

[0104] Lateral deviation refers to the vertical distance from the robot's center point to the path baseline. The path baseline is a reference line parallel to the grid lines and passing through the center of the ideal path. After image processing and coordinate transformation, the vertical distance between the robot's center point and the path baseline is the lateral deviation. This deviation value indicates the degree to which the robot deviates from the ideal path during movement. Real-time monitoring and calculation of the lateral deviation can help the robot make timely adjustments to maintain a precise travel path.

[0105] Angular deviation refers to the angle between the robot's current heading angle and the direction of the path baseline. The robot's heading angle is acquired through an inertial measurement unit (IMU), which includes attitude data such as yaw, pitch, and roll. The IMU provides the robot's orientation data in real time and calculates the heading angle. The current heading angle is compared with the direction of the path baseline to obtain the angular deviation. Angular deviation is a crucial basis for path adjustment; the robot needs to adjust its heading based on this deviation to ensure it always travels in the correct direction.

[0106] During robot operation, lateral and angular deviations need to be continuously updated at a fixed sampling period (e.g., once per second). Each update calculates new lateral and angular deviations and generates deviation trend data. This deviation data not only reflects the robot's motion state in real time but also provides a basis for subsequent dynamic path adjustments. By updating deviation trends in real time, the robot can automatically make adjustments during movement, maintaining path stability and accuracy.

[0107] This method enables the robot to accurately monitor and correct path deviations in real time. First, the use of vision sensors effectively identifies photovoltaic cell grid lines in complex environments, ensuring the accuracy of deviation calculations. Second, using heading angle data provided by the IMU, the robot can dynamically calculate the angle difference with the path baseline, further optimizing its trajectory. Furthermore, periodic deviation updates reflect the robot's motion status in real time, allowing the system to make timely path corrections and prevent the robot from deviating from the target path. Finally, the combination of these technical features effectively reduces blind spots and enables the robot to achieve high-precision path tracking during cleaning, ensuring full coverage cleaning of the photovoltaic panel surface and improving cleaning efficiency and operational reliability.

[0108] In one possible implementation, the fuzzy controller's input variables are lateral deviation and angular deviation, and its output variable is the track speed compensation value. Lateral deviation and angular deviation reflect the horizontal and angular errors in which the robot deviates from the ideal path during path tracking, respectively. The controller's goal is to adjust the speed of the left and right tracks based on these two deviations, ensuring the robot always travels along the correct path.

[0109] The fuzzy controller fuzzifies the input variables of lateral and angular deviations, dividing them into five fuzzy subsets: negative large, negative small, zero, positive small, and positive large. These subsets represent different levels of error, ranging from "negative large" indicating a significant deviation from the path to "positive large" indicating a large deviation in both direction and degree. Triangular membership functions are used to define the membership functions of these fuzzy subsets, as these functions are computationally simple and allow for precise control of the input variables.

[0110] Based on experimental calibration, a fuzzy control rule base was created. This base includes track speed compensation strategies under combined lateral and angular deviation conditions. By analyzing experimental data, a fuzzy rule based on deviations occurring during actual operation was established, specifically including the track speed compensation method when lateral and angular deviations are within a certain range. For example, if the lateral deviation is large and the angular deviation is small, a larger speed adjustment may be needed to correct the deviation.

[0111] Defuzzification is a crucial step in fuzzy control. The center-of-gravity method is used to calculate the speed compensation values ​​for the left and right tracks. This method determines the specific output value by calculating the center-of-gravity position of the fuzzy output, thus obtaining precise speed compensation values ​​for the left and right tracks. This approach transforms the fuzzy control output into a clear compensation value, which is then superimposed on the base drive speed, enabling dynamic adjustment of the robot.

[0112] Based on the deviation trend data during robot operation, the output gain coefficient of the fuzzy controller is dynamically adjusted. As the deviation changes during robot movement, the controller needs to adjust the gain coefficient in a timely manner to ensure the effectiveness of the compensation strategy. Dynamic gain adjustment can cope with different working environments and path changes, enabling the robot to maintain optimal path tracking accuracy under different conditions.

[0113] By dynamically adjusting the robot's path using fuzzy control, this invention effectively addresses complex situations encountered by the robot in practical work, such as path errors, directional deviations, and external disturbances. Compared to traditional PID control, fuzzy control better handles uncertainties and nonlinearities, exhibiting stronger robustness and adaptability. By adjusting the track speed compensation value in real time, the robot can promptly correct path deviations even with significant errors, ensuring it always travels along the predetermined path and reducing blind spots and missed areas. Furthermore, dynamically adjusting the gain coefficient allows the control system to adaptively optimize as environmental and working conditions change, thereby improving the robot's cleaning efficiency and operational stability.

[0114] In one possible implementation, the robot uses a brush track for cleaning, the ground contact length of which varies with the robot's heading angle. The heading angle is measured in real-time by an inertial measurement unit (IMU), providing attitude information such as yaw angle. Based on the current heading angle and the actual ground contact length of the brush track, the effective cleaning width can be calculated. The cleaning width refers to the actual area covered by the robot's brushes, and it is affected by the robot's travel angle. The cleaning width changes for different heading angles; therefore, real-time calculation of the cleaning width is crucial.

[0115] During the cleaning process, the robot records the cleaning effect of each segment of the path in real time, including cleaned and uncovered areas. By analyzing historical cleaning trajectory data and statistically analyzing the distribution of cleaned and uncovered areas, the cleaning effect can be accurately assessed, and potential blind spots can be identified. The distribution of uncovered areas is crucial for adjusting the cleaning path and spacing, helping the robot to better plan subsequent cleaning routes.

[0116] The minimum overlap rate threshold can be calculated based on the ratio of cleaned to uncovered areas. The overlap rate refers to the ratio of overlapping areas between adjacent cleaning paths, ensuring that uncleaned areas are covered between cleaning passes. The minimum overlap rate threshold is derived by calculating the ratio of the uncovered area to the current cleaning width. This threshold determines the overlap between two cleaning paths during the robot's next cleaning pass, ensuring comprehensive cleaning.

[0117] The robot monitors the level of contamination on the surface of photovoltaic modules in real time using visual sensors (such as cameras). The level of contamination is typically determined by the grayscale variance of the image; the higher the grayscale variance, the more severe the surface contamination. The overlap rate threshold is dynamically adjusted based on the level of contamination: when the contamination level is high, the overlap rate threshold is appropriately increased to ensure thorough cleaning; conversely, when the contamination level is low, the overlap rate can be appropriately decreased to improve work efficiency.

[0118] The overlap rate threshold calculated in real time is used as the basis for adjusting the spacing of the next cleaning cycle. Based on the current level of contamination and the areas already cleaned, the robot dynamically adjusts the spacing of the cleaning path to ensure that each cleaning operation covers the uncleaned areas while avoiding unnecessary re-cleaning. This dynamic adjustment method based on real-time data allows the robot to improve efficiency and reduce energy consumption and time waste while ensuring cleaning effectiveness.

[0119] This invention has significant advantages in path planning and blind spot elimination, which not only improves cleaning efficiency but also ensures cleaning quality, and solves the problems of path planning and uneven cleaning that robots may encounter in complex environments.

[0120] In one possible implementation, when the robot detects a continuous increase in the area of ​​uncovered regions during the cleaning process, it indicates that there are areas that are not being fully covered by the current cleaning path, possibly due to excessively large cleaning intervals. At this point, the system automatically activates an adjustment mechanism, gradually increasing the overlap rate threshold. Increasing the overlap rate threshold means that the overlap between adjacent cleaning paths increases, thereby ensuring that uncleaned areas are effectively covered in the next cleaning cycle. The increment can be set according to the size of the uncovered area and the required cleaning accuracy to avoid excessive overlap and reduce energy waste.

[0121] If the robot does not find any uncovered areas within several consecutive cleaning cycles, it indicates that the current overlap rate is sufficient and the cleaning path coverage is comprehensive. In this case, to improve cleaning efficiency and reduce repeated cleaning, the system will progressively lower the overlap rate threshold. Lowering the overlap rate threshold means reducing the overlap between adjacent paths, thereby increasing the cleaning spacing and improving cleaning efficiency. Through progressive lowering, the robot can adapt to different contamination levels and cleaning needs while maintaining cleaning quality.

[0122] To avoid excessively increasing or decreasing the overlap rate, the system sets upper and lower limits for the overlap rate threshold. These limits are determined based on the size of the photovoltaic modules and the robot's structural parameters. The size of the photovoltaic modules affects the robot's cleaning width and path planning, while the robot's structural parameters (such as the length of the brush tracks and the robot's steering capability) also influence the adjustment of the cleaning path. Setting these upper and lower limits ensures that the robot does not exceed its actual working range when adjusting the overlap rate, avoiding inefficiency due to an excessively high overlap rate or incomplete cleaning due to an excessively low overlap rate.

[0123] The alignment of the photovoltaic array significantly impacts the adjustment of the cleaning path. If the array is highly aligned, the gaps between paths are small, and the overlap rate threshold doesn't require large changes. Conversely, if the array is scattered, there may be significant unevenness between cleaning paths, necessitating a higher overlap rate to ensure thorough cleaning. By calculating the alignment of the photovoltaic array using grid line fitting error, the system can dynamically adjust the sensitivity of the overlap rate threshold. The grid line fitting error is calculated by comparing the actual and theoretical alignment of the photovoltaic array to determine the alignment, thus adjusting the overlap rate sensitivity based on the array's alignment. This allows the robot to flexibly adjust its cleaning strategy when dealing with photovoltaic arrays of varying alignment, ensuring cleaning quality.

[0124] This invention effectively improves the cleaning efficiency and quality of tracked cleaning robots through a flexible dynamic adjustment mechanism for overlap rate, and eliminates blind spots to the greatest extent possible through intelligent path planning.

[0125] In one possible implementation, a visual sensor (such as a camera or lidar) is used to capture the edge features of the photovoltaic panel. Specific methods include image processing techniques (such as edge detection algorithms) to extract the edge contour of the photovoltaic panel. By extracting the geometric parameters of the edge contour (such as length, angle, curvature, etc.), the shape, position, and relative position of the photovoltaic panel to the robot can be accurately identified. This step aids in the robot's path planning and blind spot elimination, ensuring that the robot can avoid obstacles around the photovoltaic panel and accurately align with the cleaning area.

[0126] An infrared ranging sensor at the front of the robot continuously monitors the distance to obstacles. The sensor provides minimum and average distance values, where the minimum distance represents the point closest to the obstacle, and the average distance is the distance over a certain range. This data is used to determine the presence and type of obstacles and to provide a basis for the robot's path planning. If an obstacle is too close to the robot, the system will automatically adjust the path to avoid a collision.

[0127] An inertial measurement unit (IMU) is used to monitor the robot's tilt angle changes in real time and record the rate of change. This sensor allows the robot to detect uneven or tilted surfaces, preventing instability or poor cleaning results caused by tilting. Furthermore, the data provided by the IMU can be used to adjust the robot's attitude control, ensuring it maintains the optimal cleaning angle at all times.

[0128] All sensor data will be fed into a multi-sensor data fusion decision model. The first step of this model is to assess the confidence level of each sensor's data, that is, to evaluate the reliability and validity of different sensor data. For example, visual sensors may produce errors due to changes in lighting, while infrared ranging sensors may be affected by reflection interference in certain environments. The model will assign different weights to different data based on the performance of each sensor and the current environmental conditions.

[0129] When conflicting data exists from multiple sensors, the system prioritizes the tilt angle data provided by the inertial measurement unit (IMU) as the basis for decision-making. The rate of change of tilt angle directly reflects the robot's relative position to the ground and plays a crucial role in determining the presence of obstacles (such as uneven surfaces or sloping environments). By prioritizing IMU data, the system can avoid the robot misjudging obstacles and making incorrect path planning decisions.

[0130] This invention significantly improves the accuracy and adaptability of tracked cleaning robots in path planning and obstacle identification through a multi-sensor data fusion method, and has broad application prospects.

[0131] In one possible implementation, tilt angle data comes from the robot's inertial measurement unit (IMU) to monitor the robot's tilt in real time. When the robot detects a continuous increase in tilt angle exceeding a preset safety threshold during operation, the system triggers an obstacle-crossing mode. This safety threshold is determined based on the maximum permissible tilt angle of the photovoltaic panels, aiming to prevent excessive tilting that could lead to reduced cleaning efficiency or robot tipping over. Photovoltaic panels typically have a maximum safe tilt angle; when the robot's tilt angle exceeds this range, it indicates that the robot may be approaching or crossing an obstacle (such as a step or uneven ground), necessitating the use of obstacle-crossing mode to adjust the robot's behavior.

[0132] Once obstacle-crossing mode is triggered, the control system increases the output torque of the track drive motors to provide stronger thrust, ensuring the robot can overcome obstacles. Simultaneously, the system reduces its travel speed to prevent the robot from becoming uncontrollable or unstable due to excessive speed. In this way, the robot can smoothly cross obstacles, ensuring the cleaning operation is undisturbed.

[0133] When the robot detects a sudden change in tilt angle and the infrared sensor shows a sharp increase in the distance to an obstacle, it indicates that the robot may have come into contact with a large obstacle (such as a gap between photovoltaic panels or a protrusion between panels). At this point, the system will trigger the obstacle-crossing mode. The distance information from the infrared sensor can provide changes in the distance to obstacles; when the distance increases sharply, it usually indicates that the robot has approached or crossed a relatively obvious obstacle.

[0134] In obstacle-crossing mode, the system suspends the cleaning function to prevent scratches or contamination of the photovoltaic panels when crossing obstacles. Simultaneously, the system adjusts the robot's posture to ensure smooth obstacle crossing. By controlling the differential drive of the two side tracks, the robot maintains stability while crossing obstacles, avoiding imbalance or getting stuck.

[0135] After successfully completing the obstacle-crossing or panel-spanning maneuver, the system automatically returns to normal cleaning operation mode and continues cleaning the photovoltaic panels. The recovery process is automated and requires no manual intervention, thus improving the robot's operational efficiency.

[0136] The obstacle-crossing and panel-crossing decision-making mechanism of this invention optimizes the path planning capability of the tracked cleaning robot by making reasonable use of tilt angle data and infrared sensor data. This not only improves the robot's stability and adaptability but also protects the photovoltaic panels and improves overall operation efficiency.

[0137] In one possible implementation, the robot first needs to record and store the distribution characteristics of uncovered areas during historical cleaning processes. This data is typically acquired using sensors built into the robot, such as LiDAR, infrared sensors, or image recognition systems, to mark the uncovered areas. The distribution characteristics of these uncovered areas (such as location, area, shape, etc.) are then used to generate additional cleaning paths.

[0138] The robot uses algorithms to analyze the characteristics of uncovered areas and automatically plans one or more additional cleaning paths. The planning of these additional paths is based on the actual area and shape of the uncovered areas to ensure that all blind spots are thoroughly cleaned. This process ensures that there are no dead ends or missed areas during cleaning.

[0139] To optimize cleaning efficiency and accuracy, the additional cleaning path employs a reciprocating motion pattern. This pattern efficiently covers uncleaned areas while reducing repetitive robot movement and unnecessary travel. The amplitude and direction of the reciprocating motion are determined based on the specific conditions of the uncovered area, ensuring precise alignment with its boundaries.

[0140] During reciprocating motion, the robot can adjust its range of motion in real time to ensure that it can cover all blind spots and uncleaned areas. For example, when the robot finds that there are still uncovered parts in a certain area, it can adjust its reciprocating trajectory to increase the cleaning density or range.

[0141] During the additional cleaning process, the robot monitors the cleaning effect in real time using onboard sensors (such as image recognition and infrared sensors). By comparing the current area with the preset cleanliness standard, the robot can determine whether the area has achieved the predetermined cleaning effect. If it finds that the cleanliness of a certain part has not met the standard, the robot will automatically adjust the cleaning strategy and continue cleaning until the standard is met.

[0142] The standard for cleaning completion is generally set as the removal of dirt layer in the area or the achievement of a certain brightness standard. The robot will automatically determine whether to stop the current cleaning task and move on to the next area based on the real-time detection results.

[0143] The execution results and path parameters (such as cleaning trajectory, cleaning time, and cleaning intensity) of each additional cleaning path are recorded and stored in the robot's control system. This data is not only used as a reference for subsequent cleaning tasks, but can also be analyzed using machine learning methods to help the robot optimize future path planning.

[0144] In a similar working environment, the robot can adjust its subsequent path planning strategy based on historical cleaning path parameters, thereby reducing blind spots and further improving cleaning efficiency and accuracy.

[0145] This invention, by combining the analysis of historical cleaning trajectories, the design of additional cleaning paths, real-time monitoring, and automatic adjustment, not only improves the coverage efficiency and cleaning effect of the cleaning robot, but also increases the robot's intelligence and adaptability, thereby meeting the needs for more efficient and precise cleaning.

[0146] In one possible implementation, the robot continuously collects operational data during the cleaning process, and this data is uploaded in real time to a remote monitoring center via the robot's wireless communication module (such as Wi-Fi, 4G / 5G, or LoRa). The wireless module ensures stable data transmission and real-time feedback, allowing monitoring personnel to obtain the latest progress of the cleaning operation immediately.

[0147] Uploaded content: The reported data includes:

[0148] Actual cleaning coverage: This data represents the ratio of the area actually cleaned by the robot to the total surface area of ​​the photovoltaic modules.

[0149] Uncovered area distribution map: During the cleaning process, the areas that are not covered by the robot will be marked and a distribution map will be generated to facilitate subsequent analysis and path optimization.

[0150] Obstacle handling log: Records the robot's handling process when encountering obstacles, including information such as the type and location of the obstacle, the handling method, and the handling time. These logs allow you to understand whether the robot encountered problems during execution and how to deal with them.

[0151] Coverage calculation formula: The actual cleaning coverage rate is calculated as the ratio of the area of ​​the cleaned area to the total surface area of ​​the photovoltaic modules. The area of ​​the cleaned area is confirmed by the robot's sensor system (such as vision, LiDAR, or infrared sensors), while the total surface area of ​​the photovoltaic modules is confirmed based on preset data. By comparing the area ratio of the two, the coverage efficiency of the cleaning operation is obtained.

[0152] Cleaning coverage is a key indicator for measuring the efficiency and effectiveness of cleaning operations. Through accurate coverage calculation, robots can determine whether the cleaning task has been completed and whether further adjustments to the cleaning strategy or path are needed to ensure cleaning quality.

[0153] The cleaning operation report will generate an effectiveness assessment report based on the uploaded operation data (including information on cleaning coverage, uncovered areas, and obstacle handling). The report includes the overall performance of the cleaning operation, identifies which areas were thoroughly cleaned, which areas were not fully covered, and the possible reasons (such as obstacles, improper path design, etc.).

[0154] The report also includes improvement suggestions based on the current operational status. For example, if certain areas are found to be substandard in terms of cleanliness, the report may recommend adjusting the robot's movement path or increasing the cleaning frequency of specific areas. Regarding obstacle handling, if the robot frequently encounters a certain obstacle, the report will suggest improving the obstacle recognition and avoidance strategy.

[0155] By uploading cleaning operation data to the monitoring center in real time, and combining the calculation of actual cleaning coverage, analysis of uncovered areas, recording of obstacle handling, and generation of operation reports, this invention can not only improve the transparency and efficiency of cleaning operations, but also continuously improve path planning and operation strategies through feedback and optimization mechanisms, thereby ensuring the efficient operation of the cleaning robot in complex environments.

[0156] It is understandable that:

[0157] The overlap rate threshold refers to the minimum percentage by which the cleaning widths of two adjacent cleaning paths must overlap. This threshold is the core basis for dynamically adjusting the cleaning spacing. The goal is to ensure that even with slight path deviations or terrain changes by the robot, no uncleaned blind spots are left between two cleaning paths. The threshold is not a fixed value but is dynamically calculated based on the condition of the uncovered areas and the degree of contamination on the solar panels.

[0158] Example value: The system can initially set a conservative overlap rate threshold, such as 20%. This means that the starting point of the next cleaning path will ensure a 20% overlap with the cleaning width of the previous path. If an increase in uncovered areas is detected, the system may gradually increase this threshold to 25% or 30% to increase the reliability of coverage.

[0159] The safety threshold (used for obstacle crossing decisions) refers to a preset limit value for the robot's tilt angle. When the tilt angle detected by the inertial measurement unit (IMU) continues to increase and exceeds this value, the system determines that the robot is climbing an obstacle (such as the edge of a photovoltaic panel or an uneven surface), thus triggering the obstacle crossing mode. This threshold is determined based on the maximum tilt angle that the photovoltaic module can safely withstand, preventing the robot from tipping over.

[0160] Example value: Depending on the installation structure of the photovoltaic panel and the robot's center of gravity stability, this safety threshold may be set to 15 degrees. When the robot's body tilt angle continuously exceeds 15 degrees, the obstacle-crossing mode is activated.

[0161] The minimum overlap rate threshold is a lower limit set during the dynamic calculation of the overlap rate threshold to ensure basic cleaning effectiveness. Even if the uncovered area is temporarily zero, the cleaning overlap rate cannot be lower than this value to cope with unforeseen minor deviations.

[0162] Example value: To prevent excessive reduction in overlap rate from causing missed cleaning, the system sets a minimum guaranteed value, such as 10%. This means that no matter how small the uncovered area is, the overlap rate between cleaning paths will not be lower than 10%.

[0163] A cleanliness standard is an evaluation criterion used to determine whether an area has been cleaned properly. It is typically a quantifiable indicator derived by visual sensors through the analysis of image features.

[0164] Example value: Cleanliness standard can be defined as a threshold of image grayscale variance. Before cleaning, the grayscale variance of the stained area may be large (e.g., 250); after cleaning, the clean surface has uniform grayscale, and the variance decreases. The system can be set to consider an area as having met the cleanliness standard when the image grayscale variance of a certain area is below 50.

[0165] The threshold in adaptive thresholding segmentation: In image processing, to convert a grayscale image into a black-and-white binary image to highlight raster line features, a threshold needs to be selected. Pixels with grayscale values ​​higher than this threshold are turned white, and those lower are turned black. Adaptive thresholding does not use a globally fixed threshold, but instead calculates the threshold individually based on the grayscale distribution of each small local region in the image, thus overcoming the influence of uneven illumination.

[0166] Example of value: This method itself is an algorithmic behavior, but its key parameter is the size of the local window. For example, the algorithm may divide the image into multiple 15x15 pixel small windows and calculate the threshold independently within each window (such as using the average gray level of the pixels within the window minus a constant 10 as the threshold for that window).

[0167] Output gain coefficient: In a fuzzy control system, this coefficient is used to amplify or reduce the velocity compensation value of the final output of the fuzzy inference. The system dynamically adjusts this coefficient based on the changing trends of lateral and angular deviations (e.g., whether the deviations are increasing or decreasing) to achieve a more agile or smoother control response.

[0168] Example values: When the deviation increases rapidly, the system may use a larger gain coefficient (such as 1.2) to make the robot's correction action faster; when the deviation is small and stable, a smaller gain coefficient (such as 0.8) may be used to prevent the robot from generating unnecessary jitter and make the movement smoother.

[0169] Fuzzy Controller: This is an intelligent controller, particularly suitable for complex systems where it is difficult to establish precise mathematical models. In this patent, it receives two precise input values, "lateral deviation" and "angular deviation," and then performs "fuzzification" (converting precise values ​​into fuzzy language such as "too large" or "too small"), rule-based fuzzy inference, and finally "defuzzification" (converting the fuzzy inference result into precise output values) to output the speed compensation values ​​for the left and right tracks.

[0170] Fuzzy control rule base: This is a set of language rules in the form of "IF...THEN...", which is the "brain" of the fuzzy controller. It defines which control strategy should be followed for different combinations of deviation states.

[0171] Example of a rule:

[0172] Rule 1: IF (lateral deviation is "positive") AND (angular deviation is "zero") THEN (left track speed compensation is "negative"), right track speed compensation is "positive"). / / Meaning: If the robot is severely deviated to the right but is in the correct direction, the left track speed should be significantly reduced and the right track speed increased to correct the position to the left.

[0173] Rule 2: IF (lateral deviation is "positive small") AND (angular deviation is "negative small") THEN (left track speed compensation is "negative small", right track speed compensation is "positive small"). / / Meaning: If the robot is slightly to the right and the front of the robot is slightly to the left, the track speed should be slightly adjusted for fine-tuning.

[0174] Multi-sensor data fusion decision model: This is a framework or algorithm for processing information from multiple sensors such as vision, infrared, and IMU. Its core purpose is to resolve potential conflicts between different sensor information and arrive at more reliable environmental perception conclusions. The model first evaluates the reliability (confidence level) of the data from each sensor.

[0175] Confidence assessment: This is a key step in the multi-sensor data fusion model, used to quantify the reliability of each sensor in the current environment.

[0176] Evaluation Example: In strong, reflective light environments, the confidence level of visual sensors in identifying edges may be reduced (e.g., assigned a weight of 0.4), while the confidence level of infrared ranging sensors, which are less affected by lighting, remains high (weight 0.7). The tilt angle data provided by the IMU is the most stable and has the highest confidence level (weight 0.9). When data conflicts occur, the system will prioritize information from the high-confidence sensor.

[0177] Upper and lower limits (for overlap rate threshold): To prevent the overlap rate threshold from getting out of control during dynamic adjustment (too high leads to inefficiency, too low leads to missed washing), the allowed range of variation is set for it.

[0178] Example of value: Based on the size of the photovoltaic panel (e.g., the width of the module is 1 meter) and the cleaning width of the robot (e.g., 0.8 meters), the lower limit of the overlap rate threshold can be set to 5% (to ensure minimum coverage) and the upper limit can be set to 40% (to avoid over-cleaning).

[0179] Alignment regularity describes the degree of regularity in the arrangement of components within a photovoltaic array. The less regular the arrangement, the greater the challenge for path planning.

[0180] Calculation example: The average error between the grid lines fitted by the Hough transform and the least squares method and the ideal standard straight line can characterize the neatness. For example, an average fitting error of less than 5 mm can be considered "neat", an error between 5 and 20 mm is "average", and an error greater than 20 mm is "unnecessary".

[0181] The grid line fitting error is the average distance between all detected grid line pixels and the fitted ideal straight line when the least squares method is used to fit the detected grid lines. This error value is directly used to calculate the "alignment neatness".

[0182] Image grayscale variance: In image processing, variance measures the dispersion of pixel grayscale values ​​in an image. A clean photovoltaic panel surface exhibits uniform grayscale with low variance; while areas contaminated by dust, bird droppings, etc., show significant grayscale variation and high variance. Therefore, this value can serve as a quantitative indicator of the degree of contamination.

[0183] Example values: The grayscale variance of a clean photovoltaic panel image may be about 30; while the grayscale variance of a heavily polluted area may reach more than 200.

[0184] Expected overlap rate: The overlap rate that the system expects to achieve before starting a cleaning cycle, based on historical data and the current status. It is the target value for dynamically adjusting the cleaning interval.

[0185] Effective cleaning width: The actual cleaning width covered by the robot's brushes. It is not fixed and equal to the physical width of the brushes, but varies with the robot's heading angle (the angle between the direction of travel and the direction of the grid lines). The larger the heading angle, the narrower the effective cleaning width.

[0186] Calculation example: If the physical width of the brush is W and the current heading angle is θ, then the effective cleaning width ≈ Wcos(θ). When θ = 0 degrees (completely parallel to the grid lines), the cleaning width is the maximum, which is W; when θ = 10 degrees, the cleaning width is slightly less than W.

[0187] Lateral deviation and angular deviation:

[0188] Lateral deviation: The vertical distance from the robot's center point to the preset path baseline. Units are millimeters (mm) or centimeters (cm). Positive values ​​may indicate a deviation to the right, and negative values ​​to the left.

[0189] Angular deviation: The angle between the robot's current heading and the path baseline. The unit is degrees (°). A positive value may indicate the robot is veering to the right, and a negative value may indicate the robot is veering to the left.

[0190] Fixed sampling period: The control system samples and calculates sensor data at fixed time intervals to update the deviation value.

[0191] Example of a value: In order to ensure the real-time performance of the control, the sampling period is usually short, such as once every 100 milliseconds (0.1 seconds).

[0192] Deviation trend data: This is a numerical sequence of lateral and angular deviations over a period of time (such as the most recent samplings). By analyzing this sequence, it can be determined whether the deviation is increasing, decreasing, or remaining stable, thereby predicting the robot's motion trend and using it for forward control (such as adjusting the gain coefficient of fuzzy control).

[0193] Contamination level: A quantitative description of the dirtiness on the surface of photovoltaic modules, usually reflected by indicators such as "image grayscale variance".

[0194] Actual cleaning coverage: A key performance indicator (KPI) for evaluating the completion of a cleaning operation. The calculation formula is: Actual cleaning coverage = (Area of ​​cleaned area / Total surface area of ​​photovoltaic modules) × 100%.

[0195] The following examples will illustrate this in detail:

[0196] This invention aims to solve the problem of accurate path tracking in photovoltaic cleaning robots, especially in complex environments. It addresses how to utilize the robot's vision sensors and inertial measurement unit (IMU) to monitor and adjust the robot's path to ensure efficient and accurate cleaning. This invention achieves precise grid line detection and path adjustment by introducing lateral deviation, angular deviation, Hough transform, and least squares fitting techniques, thereby improving robot cleaning efficiency and operational safety.

[0197] Lateral deviation represents the vertical deviation of the robot from the ideal path, while angular deviation represents the angular difference between the robot's heading and the path baseline. Specifically, the robot obtains its heading angle through an IMU, identifies the path baseline through a vision sensor, identifies the photovoltaic cell grid lines through a Hough transform, and finally calculates the deviation of the robot's current position relative to the ideal path.

[0198] The effective cleaning width varies with the robot's heading angle, and the calculation formula is as follows:

[0199] ;

[0200] Where W is the physical width of the brush and θ is the robot's heading angle.

[0201] The robot updates lateral and angular deviations at a fixed sampling period (e.g., 100 milliseconds) and generates deviation trend data. By analyzing this trend data, the robot can predict future path deviations and make adjustments in advance to maintain path stability.

[0202] The degree of contamination is reflected by the image grayscale variance. The actual cleaning coverage is calculated using the following formula:

[0203] ;

[0204] Suppose there is an area of ​​20m 2 The robot needs to perform a full-coverage cleaning of the photovoltaic panel array. The robot is equipped with the following hardware:

[0205] Visual sensor: Used to capture image information and identify the grid lines of photovoltaic cells.

[0206] IMU (Inertial Measurement Unit): Used to obtain the robot's heading angle.

[0207] LiDAR or infrared sensors: used to record uncleaned areas.

[0208] Initial setup of the robot:

[0209] Brush physical width: 1m;

[0210] Initial heading angle: 0° (parallel to grid lines);

[0211] Initial cleaning spacing: 1m;

[0212] Deviation sampling period: 100 milliseconds;

[0213] Contamination threshold: An image grayscale variance greater than 50 indicates severe contamination and requires cleaning;

[0214] The robot initially moves along the grid lines at a 0° heading angle, parallel to the grid lines, with an initial cleaning width of 1m (brush width). Every 100 milliseconds, the IMU sensor updates the heading angle, and the vision sensor detects the photovoltaic cell grid lines and calculates the lateral deviation.

[0215] Assuming the robot experiences a 2° heading deviation during its movement, the effective cleaning width becomes:

[0216] ;

[0217] As can be seen, the change in heading angle causes a slight change in the effective cleaning width, but this change is small enough not to affect the cleaning quality.

[0218] When the robot's lateral deviation reaches a certain value (e.g., greater than 10cm), the robot predicts future deviations based on deviation trend data and makes path corrections in advance. For example, when the lateral deviation tends to increase, the robot will adjust its direction of travel accordingly, ensuring it moves along the grid lines. This ensures that the cleaning coverage is not affected by the deviation.

[0219] In a cleaning task, the robot detects uncleaned areas (such as areas where dirt has not been completely removed) using LiDAR or infrared sensors. Based on this data, the robot plans additional cleaning paths and uses a reciprocating motion pattern to clean. Assume the total area of ​​the uncleaned area is 5m². 2 The robot will generate additional paths based on these areas to ensure that all areas are thoroughly cleaned.

[0220] After cleaning, the robot uses image recognition or infrared sensors to evaluate the cleaned area and determine whether the predetermined cleaning effect has been achieved. If the cleanliness of a certain area does not meet the standard, the robot will automatically adjust the cleaning strategy, increasing the cleaning density or changing the cleaning method until the area meets the cleaning standard.

[0221] Assuming the robot successfully cleaned 15m 2 The actual cleaning coverage rate of the photovoltaic panel area is:

[0222] Actual cleaning coverage =

[0223] This coverage rate is a key performance indicator representing the completion of the cleaning task, reflecting the efficiency of the robot in the cleaning process.

[0224] Traditional photovoltaic cleaning methods often rely on manual operation, with relatively simple robot path planning and a fixed cleaning width. Due to the lack of real-time deviation monitoring and adjustment mechanisms, robots are prone to deviating from the ideal path, resulting in incomplete cleaning and even leaving blind spots. Furthermore, traditional methods struggle to dynamically adjust the cleaning path based on actual conditions, leading to low efficiency and wasted resources.

[0225] This invention, by calculating lateral and angular deviations in real time, can accurately monitor the robot's motion state and adjust the path promptly, avoiding the deviation and omission problems common in traditional methods. Simultaneously, by employing Hough transform and least squares fitting techniques, the accuracy of grid line extraction is significantly improved, making path planning more reliable. Furthermore, by dynamically adjusting the cleaning spacing and planning additional cleaning paths, the robot can achieve full-coverage cleaning, improving cleaning efficiency and quality.

[0226] Through actual testing and comparison with traditional cleaning methods, the robot using this invention significantly improves cleaning efficiency by increasing the cleaning coverage from 60% to over 75% within the same cleaning time. Simultaneously, the robot reduces blind spots and missed areas through real-time path correction and additional path planning, ensuring comprehensive cleaning results.

[0227] Through specific application scenarios described in the above embodiments, and by combining techniques such as lateral deviation, angular deviation, Hough transform, and least squares fitting, this invention achieves highly efficient cleaning of photovoltaic panels. It not only improves cleaning coverage and efficiency but also automatically adjusts the path to avoid deviations and blind spots, thus enhancing operational safety and stability. The implementation of this invention can significantly improve the application effect of robots in the field of photovoltaic cleaning and possesses strong market competitiveness.

[0228] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0229] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for path planning and blind area elimination of a track cleaning robot, characterized in that, Includes the following steps: Images of the photovoltaic module surface are acquired by a vision sensor installed on the robot body, the grid line features of the photovoltaic cells are extracted, and a local coordinate system is established. A path baseline is generated based on the grid line direction, and the lateral and angular deviations between the robot's current position and the path baseline are calculated in real time. Based on the lateral and angular deviations, the robot's walking path is dynamically adjusted using fuzzy control. A map of the cleaned area is generated based on historical cleaning trajectory records, and the cleaning spacing is dynamically adjusted by calculating the current cleaning width and the expected overlap rate. Obstacle detection is performed by fusing data from multiple sensors, and when sensor data conflict, obstacle crossing or board crossing decisions are made based on tilt angle data. The specific method for dynamically adjusting the cleaning spacing by calculating the current cleaning width and the expected overlap rate includes: The effective cleaning width is calculated based on the actual ground contact length of the brush track and the robot's current heading angle, which is measured in real time by an inertial measurement unit; By analyzing historical cleaning trajectory data, the distribution of cleaned and uncovered areas can be statistically analyzed. The minimum overlap threshold required is calculated based on the ratio of the uncovered area to the current cleaning width; The overlap rate threshold is dynamically adjusted based on the degree of surface contamination of the photovoltaic module. The degree of contamination is determined by the image grayscale variance analyzed by a visual sensor. The overlap rate threshold calculated in real time is used as the basis for adjusting the spacing of the next cleaning cycle; The dynamic adjustment method for the overlap rate threshold also includes: When the area of ​​the uncovered region is detected to be continuously increasing, the overlap rate threshold is increased in a stepwise manner. When no uncovered areas are found in multiple consecutive cleaning cycles, the overlap rate threshold is gradually reduced. Set upper and lower limits for the overlap rate threshold, which are determined based on the size of the photovoltaic module and the robot's structural parameters; The sensitivity of the overlap rate threshold is adjusted based on the alignment of the photovoltaic array, and the alignment is calculated by the grid line fitting error.

2. The path planning and blind area elimination method for a track cleaning robot according to claim 1, wherein, The specific steps for extracting the grid line features of photovoltaic cells and establishing a local coordinate system include: The color images acquired by the vision sensor are converted to grayscale, transforming the RGB color space into a grayscale space. An adaptive threshold segmentation method is used to binarize grayscale images. The threshold of the adaptive threshold segmentation method is dynamically determined based on the grayscale distribution characteristics of local regions of the image. The grid lines of photovoltaic cells are identified by detecting straight line features in a binary image using Hough transform. The least squares method is used to fit the equations of the detected grid lines, and the coordinates of the intersection points between the grid lines are calculated. Choose the intersection of the grid lines closest to the image center as the origin of the local coordinate system, and use the direction of grid line extension as the axis of the coordinate system.

3. The path planning and blind area elimination method for a track cleaning robot according to claim 1, wherein, The specific methods for real-time calculation of the lateral and angular deviations between the robot's current position and the path baseline include: The robot's current position image is acquired in real time using a vision sensor, and the offset of the current position relative to the grid lines is identified. The vertical distance from the robot's center point to the path baseline is calculated as the lateral deviation. The path baseline is a reference line that is parallel to the grid lines and passes through the center of the ideal path. The angle between the robot's current heading angle and the path baseline direction is calculated as the angle deviation. The heading angle is obtained through attitude data provided by the inertial measurement unit. The values ​​of lateral deviation and angular deviation are continuously updated at a fixed sampling period to form deviation trend data.

4. The path planning and blind area elimination method for a track cleaning robot according to claim 1, wherein, The specific implementation method of dynamically adjusting the robot's walking path through fuzzy control includes: A fuzzy controller is established with lateral deviation and angular deviation as input variables and track speed compensation value as output variable; The fuzzy subsets of the input variables are defined to include five levels: negative large, negative small, zero, positive small, and positive large. The membership function of each fuzzy subset adopts a triangular function. A fuzzy control rule base is established based on experimental calibration. The fuzzy control rule base includes a track speed compensation strategy based on the combination of lateral deviation and angular deviation. The accurate left and right track speed compensation values ​​are obtained by defuzzification calculation using the center of gravity method, and then the left and right track speed compensation values ​​are superimposed on the basic drive speed. The output gain coefficient of the fuzzy controller is dynamically adjusted based on the deviation change trend data.

5. The method for path planning and blind spot elimination of a tracked cleaning robot according to claim 1, characterized in that, The specific steps for obstacle identification by fusing multi-sensor data include: The edge features of the photovoltaic panel are identified by a visual sensor, and the geometric parameters of the edge contour are extracted. The distance information of obstacles in front of the robot is detected by an infrared ranging sensor, and the distance information includes a minimum distance value and an average distance value. The robot's tilt angle is monitored in real time using an inertial measurement unit, and the rate of tilt angle change is recorded. A multi-sensor data fusion decision model is established, which first evaluates the confidence level of each sensor's data. When there are conflicting results from different sensors, the tilt angle data provided by the inertial measurement unit shall be given priority as the basis for decision-making.

6. The method of claim 5, wherein, The specific methods for making obstacle-crossing or board-crossing decisions based on tilt angle data include: When the tilt angle is detected to be continuously increasing and exceeding the safety threshold, the obstacle crossing mode is triggered. The safety threshold is determined based on the maximum allowable tilt angle of the photovoltaic module. In obstacle-crossing mode, the control system increases the output torque of the track drive motor while reducing the travel speed; When a sudden change in tilt angle is detected and the distance value displayed by the infrared sensor increases sharply, the cross-board mode is triggered; In cross-plate mode, the cleaning function is paused and the body posture is adjusted. Stable cross-plate movement is achieved by controlling the differential speed of the two side tracks. After completing the obstacle crossing or board crossing action, it automatically resumes the normal cleaning operation mode.

7. The path planning and blind area elimination method for a track cleaning robot according to claim 1, wherein, The process includes blind spot compensation after dynamically adjusting the cleaning spacing. Additional cleaning paths are generated based on the distribution characteristics of uncovered areas in historical cleaning trajectories. The additional cleaning path adopts a reciprocating motion mode, and the motion range is determined according to the area and shape of the uncovered area; During the execution of the additional cleaning path, the cleaning coverage effect is monitored in real time until the predetermined cleanliness standard is reached.

8. The method of claim 1, wherein, The process of making an obstacle-crossing or board-crossing decision also includes a data reporting step: The cleaning operation data is uploaded to the monitoring center via a wireless communication module; The cleaning operation data includes the actual cleaning coverage rate, a map of uncovered areas, and an obstacle handling log. The actual cleaning coverage rate is calculated as the ratio of the cleaned area to the total surface area of ​​the photovoltaic modules. A cleaning operation report is generated, including performance evaluation and improvement suggestions for this operation.

Images