Intelligent control method of double-sucker cleaning robot adaptive to multi-form glass curtain wall

Abstract

The invention belongs to the technical field of glass cleaning robots, and discloses an intelligent control method of a double-sucker cleaning robot adaptive to a multi-form glass curtain wall, which comprises the following steps of: determining a fixed anchor point; double vacuum suction cups in the double-suction-cup cleaning robot are ensured to be in a stable suction state; state sensing: after the fixed anchor point is determined in the step S1, starting a state sensing system, synchronously completing environment and obstacle sensing and self attitude sensing, collecting and processing multi-source sensing data, and providing accurate input for subsequent control strategy decision making; intelligent obstacle avoidance: after the posture position and the obstacle feature information of the double-sucker cleaning robot are obtained in the step S2, planning and controlling the motion path of the robot, so that the robot can cross the front obstacle and is adsorbed to the next position; through multi-source sensing fusion and a dynamic control strategy, precise adaptation and stable operation of multi-form curtain walls such as planar curtain walls, arc curtain walls and curtain walls with protruding obstacles are achieved.

Description

Technical Field

[0001] This invention relates to the field of glass cleaning robot technology, and specifically to an intelligent control method for a dual-suction cup cleaning robot adapted to various types of glass curtain walls. Background Technology

[0002] With the rapid development of the modern construction industry, glass curtain walls, due to their advantages such as good lighting and aesthetic appeal, are widely used in the facade design of high-rise buildings. This has led to a huge demand for curtain wall cleaning. Traditional manual cleaning methods are not only inefficient and labor-intensive, but also pose serious safety risks such as falls from heights, failing to meet the large-scale, routine cleaning needs of high-rise building curtain walls. Therefore, automated wall-climbing cleaning robots have become a core solution to replace manual labor. Among various wall-climbing cleaning robots, the dual-suction cup robot has become the mainstream technology due to its compact structure, strong suction stability, and high climbing efficiency. Its core working principle is to use a cyclical adsorption-movement pattern of two vacuum suction cups, combined with a telescopic and rotary transmission mechanism, to achieve wall climbing. Simultaneously, an integrated cleaning module completes the work concurrently, effectively solving the safety and efficiency pain points of manual cleaning. It can be widely used in cleaning flat glass curtain walls.

[0003] However, with the diversification of architectural design, glass curtain walls have gradually evolved from traditional single planar forms to irregular and complex shapes such as curved, zigzag, and those with protruding obstacles like keels, joints, and embedded parts. Existing dual-suction-cup cleaning robot control technologies are mostly based on fixed parameter designs, optimized only for flat planar curtain walls, lacking the ability to adapt to irregular curtain wall shapes. This fails to meet the complex operational needs of irregularly shaped glass curtain walls, necessitating an intelligent control method that can accurately adapt to various curtain wall shapes and dynamically adjust control parameters to expand the robot's application scenarios and improve operational stability and reliability in complex environments. Summary of the Invention

[0004] To address the shortcomings of existing dual-suction cup cleaning robot control technology, which is only compatible with flat curtain walls and lacks adaptive capability for non-frameless curtain walls, this method is based on a robot multimodal perception-control integrated system. Through multi-source sensor fusion and dynamic control strategies, it achieves accurate adaptation and stable operation of curtain walls of various shapes, including flat, curved, and those with protruding obstacles.

[0005] To achieve the above objectives, according to one aspect of the present invention, a smart control method for a dual-suction cup cleaning robot adapted to various types of glass curtain walls is provided, comprising the following steps: S1 Fixed Anchor Point Establishment: By combining the vacuum degree sensor and pressure sensor on the vacuum suction cup, the dual vacuum suction cups in the dual suction cup cleaning robot are ensured to be in a stable adsorption state, thereby establishing the fixed anchor point; S2 State Awareness: After establishing the fixed anchor point in step S1, the state awareness system is activated to simultaneously complete the perception of the environment and obstacles as well as the perception of its own posture, collect and process multi-source sensor data, and provide accurate input for subsequent control strategy decisions; S3 Intelligent Obstacle Avoidance: After obtaining the robot's own posture and position and obstacle feature information in step S2, the robot's movement path is planned and controlled so that the robot can cross the obstacles in front and attach to the next position.

[0006] Preferably, in step S1, the vacuum sensor monitors the vacuum pressure inside the suction cup in real time, and the pressure sensor detects the adhesion pressure between the suction cup's adsorption surface and the wall. When the vacuum pressure detected by the vacuum sensor is ≥0.09MPa, and the adhesion pressure detected by the pressure sensor is stable in the range of 0.01-0.02MPa, it is determined that the suction cup is firmly adsorbed and the fixed anchor point is effectively established.

[0007] Preferably, step S2 specifically includes the following steps: S21 Robot Attitude Perception: Relevant data is acquired through an IMU integrated above the center of the cleaning robot's horizontal arm. After preprocessing the raw data, the robot's roll angle is calculated. Pitch angle Yaw angle The robot's posture calculations are fused and corrected, and the rotation angles and speeds of the motors recorded by the joint motor encoders are combined to finally output the posture parameters of each part of the robot. S22 Environmental Obstacle Perception: Through binocular cameras and ultrasonic sensors, obstacles in front are calculated to obtain key parameters such as obstacle height, obstacle thickness, and obstacle distance.

[0008] Preferably, in step S21, the IMU includes a triaxial accelerometer and a triaxial gyroscope, and the relevant data it acquires includes triaxial acceleration. With triaxial angular velocity ; The above data preprocessing method is as follows: A moving average filter is used for preprocessing, specifically, the obtained data including triaxial acceleration is preprocessed... With triaxial angular velocity The sensor data X, with Size of the filter window The sampled data at time t is preprocessed, and the formula for the preprocessed data is:

[0009] Where: X represents the raw sensor data and ; At the current sampling time, This is the size of the filtering window; This is the filtered data.

[0010] As a preferred option, the accelerometer and gyroscope data calculation method in step S21 is as follows: complementary filtering is used to fuse accelerometer and gyroscope data, the roll angle and pitch angle are roughly calculated using the gravity component of the accelerometer, and then the attitude change is predicted based on the integral of the gyroscope angular velocity.

[0011] As a preferred embodiment, the fusion and correction method in step S21 is as follows: using the yaw angle calibration value obtained by the binocular camera, the value is updated every 5 seconds through binocular camera visual feature matching, and the drift is corrected by adaptive fusion weights with the acceleration value; the aforementioned adaptive fusion weights are obtained from the vibration intensity, the vibration intensity is evaluated based on the acceleration data, different vibration levels are divided, and the corresponding adaptive fusion weights are obtained according to the vibration level.

[0012] As a preferred option, the specific steps of step S22 are as follows: S221: The binocular camera is calibrated, and distortion correction and grayscale conversion are performed on the left image, followed by a smooth grayscale image after Gaussian filtering; the Sobel operator is used to calculate the gradient and output the edge enhancement image; the global grayscale value of the edge enhancement image is calculated, and an adaptive thresholding strategy is used to extract obstacle edges, output a binary edge image, perform dilation operation to fill small gaps in the edges and eliminate internal holes in the edges, and output a complete contour image; S222: Use the eight-neighbor labeling method to label connected regions, traverse the complete contour map output in step S221, label all connected regions, and output the region set; S223: Output the mask image of the only valid obstacle region after filtering, and extract 3 core feature points by combining the upper surface contour; S224: Using the SGBM algorithm, only the above 3 feature points are matched in the right image of the binoculars to calculate the obstacle height, obstacle distance and obstacle thickness; S225: The information acquired by the ultrasonic sensor is fused with the information from the camera to correct the obstacle height and obstacle distance obtained in step S224.

[0013] Preferably, step S3 includes both accessible and accessible scenarios. For barrier-free scenarios, it adopts a normal gait, with dual suction cups alternately adsorbing, translating, and rotating, and the rotating motor follows a pre-planned S-shaped curve motion trajectory. For obstacle-prone scenarios, an obstacle-crossing gait is adopted. The suction cup lifting angle, translation distance and adsorption gap are dynamically adjusted through a closed-loop algorithm. Combined with the master-slave motor synchronous control strategy, filtering, feedforward compensation and triple feedback correction are superimposed to suppress jitter. Anomaly adaptive logic is also integrated to adjust the control parameters in real time when the curtain wall shape changes abruptly.

[0014] Preferably, the S-shaped curve motion trajectory is the result of the rotary motor rotating to an angle. The complete trajectory cycle consists of an acceleration phase, a deceleration phase, a constant speed phase, an acceleration / deceleration phase, and a deceleration / deceleration phase.

[0015] As a preferred embodiment, the closed-loop algorithm is as follows: based on the obstacle parameters under environmental visual perception and real-time ultrasonic obstacle perception, after raising to the obstacle height and the ultrasonic waves no longer sensing the obstacle, raise to a reserved safe height, set the rotation angle according to the distance to the obstacle and the obstacle thickness, and perform obstacle-crossing rotation action. The master-slave motor synchronization control strategy is as follows: a master-slave architecture is adopted to perform synchronous control of the two motors. The slave motor tracks the actual angle of the master motor and plans the control command for the next moment based on the actual angle, and performs synchronization correction and angular velocity jitter and crossarm rotation jitter suppression correction. The slave motor receives the basic trajectory command and rotates. When receiving the command, a low-pass filter is used to filter out command noise higher than the jitter frequency band, and the load fluctuation compensation command fed back by the current sensor is used to offset the jitter caused by load changes due to robot movement. When the initial posture is not parallel, under the condition of adsorption and stabilization on one side, the corresponding vertical arm on the other side is raised, and the crossarm is rotated to correct the horizontal position. Then, the vertical arms are raised in sequence to correct the motor rotation angle.

[0016] In summary, the technical solutions conceived by this invention have the following beneficial effects compared with the prior art: The method of this invention effectively breaks through the application limitations of traditional control methods, significantly improves the operational stability, reliability and adaptability of dual suction cup cleaning robots under complex curtain walls, expands the application scenarios of automated cleaning robots, meets the large-scale and routine cleaning needs of multi-form glass curtain walls in high-rise buildings, and completely solves the industry pain points of low efficiency and high safety risks of manual cleaning. Attached Figure Description

[0017] Figure 1 This is a perspective view of the dual-suction cup glass curtain wall cleaning robot involved in this application.

[0018] Figure 2 This is a flowchart of the intelligent control algorithm for the dual-suction cup glass curtain wall cleaning robot involved in this application.

[0019] Figure 3 The mechanical design of the dual-suction cup glass curtain wall cleaning robot involved in this application is simplified. Figure 1 .

[0020] Figure 4 The mechanical design of the dual-suction cup glass curtain wall cleaning robot involved in this application is simplified. Figure 2 .

[0021] Figure 5 The figure shown is a schematic diagram of the angular velocity and angular acceleration curves of the rotating electric motor involved in this application. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0023] Please see Figure 1 and Figure 2 This invention provides an intelligent control method for a dual-suction cup cleaning robot adapted to various types of glass curtain walls. Its main improvements are as follows: This method is based on a multimodal perception-control integrated system for dual-suction cup glass curtain wall cleaning robots. A multi-type curtain wall adaptation perception module is equipped on the dual-suction cup wall-climbing cleaning robot. An ultrasonic distance sensor detects the wall spacing and obstacle boundaries of different curtain wall shapes (planar, curved, and with protruding obstacles). Combined with a binocular camera, the curtain wall shape contour is extracted and dimensional deviations are calibrated, thereby achieving accurate quantitative assessment of obstacles and wall state recognition for various curtain wall shapes. The system is equipped with a multi-type adaptation intelligent control module. A motor encoder acquires multi-axis pose data of the robot, and a closed-loop control algorithm dynamically adjusts the lifting angle, lifting distance, and suction gap of the dual suction cups. This optimizes obstacle-crossing efficiency for planar curtain walls, adapts pose compensation for curved curtain walls, and strengthens landing point safety verification for curtain walls with obstacles. Simultaneously, it integrates anomaly adaptive logic to automatically adjust control parameters when the curtain wall shape changes abruptly, ensuring stable climbing and operation of the cleaning robot under various curtain wall shapes.

[0024] See Figure 1 This is a three-dimensional view of the dual-suction cup glass curtain wall cleaning robot provided by the present invention.

[0025] According to an embodiment of the present invention, the dual-suction cup cleaning robot adapted to multi-shaped glass curtain walls includes a mechanical transmission module, a cleaning module, a multimodal perception and decision-making module, and a power supply module. The mechanical transmission module comprises three parts: a horizontal arm, a vertical arm, and a dual vacuum suction cup assembly.

[0026] According to one embodiment of the present invention, the two horizontal arms and the vertical arm are connected by a motor-driven gear slider. The vertical arm can be controlled by the motor gear system to be raised in the vertical direction by the guide rail constraint. The dual vacuum suction cup assembly is located below the vertical arm and is connected to the deep groove ball bearing integrated joint by a rotary motor. The vacuum suction cup assembly can be controlled by the motor to rotate around the vertical direction, and can also rotate freely around the direction parallel to the plane within a certain limit.

[0027] According to one embodiment of the present invention, the cleaning module includes two cleaning brushes, which are respectively connected to two suction cup assemblies via a rotary motor. The cleaning function can be achieved by rotating the brushes in a vertical direction under motor control.

[0028] According to one embodiment of the present invention, the multimodal sensing and decision-making module includes a binocular camera and an IMU, an ultrasonic sensor, a vacuum sensor, a pressure sensor, and a data processing and decision-making unit. The binocular camera and IMU are located above the center of the horizontal arm; the ultrasonic sensors are respectively located at the bottom of the two vertical arms and above the vacuum suction cup assembly; the vacuum sensor and pressure sensor are respectively located at the edges of the suction surfaces of the two suction cups; the data processing and decision-making unit is located inside the horizontal arm and is connected to the binocular camera and IMU, the ultrasonic sensor, the vacuum sensor, and the pressure sensor via data connection lines, and is also connected to each motor via data connection lines, for receiving and outputting sensing information and control signals.

[0029] According to one embodiment of the present invention, the power module is located inside the cross arm and is mainly a rechargeable lithium battery, which is connected to the motor, binocular camera, data processing and decision-making unit through a power line.

[0030] See Figure 2 The flowchart illustrates an intelligent control method for a dual-suction cup cleaning robot adapted to various glass curtain wall shapes, comprising the following steps: To achieve cleaning operations on the glass curtain wall of a building, this invention provides a method for cleaning a dual-suction cup cleaning robot adapted to various glass curtain wall shapes. S1 Fixed Anchor Point Establishment: By combining the vacuum degree sensor and pressure sensor on the vacuum suction cup, the dual vacuum suction cups in the dual suction cup cleaning robot are ensured to be in a stable adsorption state, thereby establishing the fixed anchor point; The robot's dual vacuum suction cups adsorb onto the wall, forming a fixed anchor point for operation and providing a stable reference for subsequent perception and action execution.

[0031] In this embodiment, the adhesion degree of the vacuum suction cup is jointly measured by a vacuum sensor and a pressure sensor. The vacuum sensor monitors the internal vacuum pressure of the suction cup in real time, while the pressure sensor detects the adhesion pressure between the suction cup's adhesion surface and the wall surface. When the vacuum pressure detected by the vacuum sensor is ≥0.09 MPa, and the adhesion pressure detected by the pressure sensor is stable within the range of 0.01-0.02 MPa, the suction cup is considered to be firmly adhered, and the anchor point is considered effectively established.

[0032] S2 State Awareness: After establishing the fixed anchor point in step S1, the state awareness system is activated to simultaneously complete the perception of the environment and obstacles as well as the perception of its own posture, collect and process multi-source sensor data, and provide accurate input for subsequent control strategy decisions; The specific process includes: S21 Robot Attitude Perception: Relevant data is acquired through an IMU integrated above the center of the cleaning robot's horizontal arm. After preprocessing the raw data, the robot's roll angle is calculated. Pitch angle Yaw angle The robot's posture calculations are fused and corrected, and the rotation angles and speeds of the motors recorded by the joint motor encoders are combined to finally output the posture parameters of each part of the robot. In this embodiment, the IMU used includes a three-axis accelerometer and a three-axis gyroscope, and the relevant data acquired includes three-axis acceleration. With triaxial angular velocity That is, during data preprocessing, the three-axis acceleration in the Earth coordinate system will be input. With triaxial angular velocity The moving average filter preprocessing is used, specifically to preprocess the obtained triaxial acceleration. With triaxial angular velocity The sensor data X, with Size of the filter window The sampled data at time t is preprocessed, and the formula for the preprocessed data is:

[0033] Where X represents the raw sensor data. and ; At the current sampling time, This is the size of the filtering window; This is the filtered data. In this embodiment, a 5-frame window is used for preprocessing, i.e. Pick .

[0034] In this embodiment, the robot's attitude calculation employs complementary filtering to fuse accelerometer and gyroscope data. During the calculation, considering the robot's smooth motion during operation, the influence of linear acceleration is ignored, and acceleration is utilized instead. The gravitational components are used to roughly calculate the roll and pitch angles, i.e.:

[0035]

[0036] in, , These are the roll angle and pitch angle, roughly calculated from the accelerometer. The attitude change is predicted based on the integral of the gyroscope angular velocity:

[0037] ,

[0038] in: , , These are the roll, pitch, and yaw angles predicted by the gyroscope at the current moment. , , The attitude angles calculated at the previous moment; The sampling period is matched with the IMU update frequency. In this embodiment, the update frequency is 100 Hz, i.e. .

[0039] In this embodiment, the fused and corrected attitude angles from the attitude calculation are as follows:

[0040]

[0041]

[0042] in, This is the yaw angle calibration value; The fusion and correction method is as follows: using the yaw angle calibration value obtained by the binocular camera, it is updated every 5 seconds through binocular camera visual feature matching, and the drift is corrected by adaptive fusion weights with the acceleration value. The aforementioned adaptive fusion weights are obtained from the vibration intensity. The vibration intensity is evaluated based on the acceleration data, different vibration levels are divided, and the corresponding adaptive fusion weights are obtained according to the vibration level. To achieve adaptive fusion weights, the weights are adjusted based on vibration intensity; the more intense the vibration, the lower the weight of the gyroscope. The smaller.

[0043] In this embodiment, vibration intensity is evaluated based on acceleration data. The triaxial accelerations after moving average filtering preprocessing are used. The combined acceleration is .

[0044] The formula for calculating the short-term standard deviation of the most recent M-frame data is:

[0045] in, for The average value of adjacent M frames of data is used; in this embodiment, M is set to 10. The high-frequency component energy is calculated, and the high-frequency components of acceleration are extracted using differential operations to quantize burst interference. .

[0046] Based on the actual measurement data of the robot operation, a three-level vibration standard was set.

[0047]

[0048] In this embodiment, see Figure 3, 4 The simplified mechanical diagram involves a dual-suction cup glass curtain wall cleaning robot, with a horizontal arm length of... The length from the bottom of the vertical arm to the end of the guide rail is... The vertical arm guide rail restricts vertical movement, with a limit value of [value missing]. The end-rotor motor limits the rotational motion, with the limit value being... The encoder records the vertical lifting amount of the two vertical arm crossbars along the straight bar direction. and Rotary motor rotation angle and The sensor at the center of the horizontal arm is set as the origin, the direction of the horizontal arm is the x-axis, the direction perpendicular to the horizontal arm and parallel to the glass plane of the curtain wall is the y-axis, and the direction perpendicular to the glass plane of the curtain wall is the z-axis. This forms the robot coordinate system, with the relative positions of the two end effectors as follows: The posture is , In this embodiment, a dual-suction cup glass curtain wall cleaning robot... , , .

[0049] S22 Environmental Obstacle Perception: Using a binocular camera and ultrasonic sensors, obstacles in front are analyzed to obtain quantitative core parameters such as obstacle height, obstacle thickness, and obstacle distance. The specific steps are as follows: S221: The binocular camera is calibrated, and distortion correction and grayscale conversion are performed on the left image, followed by a smooth grayscale image after Gaussian filtering; the Sobel operator is used to calculate the gradient and output the edge enhancement image; the global grayscale value of the edge enhancement image is calculated, and an adaptive thresholding strategy is used to extract obstacle edges, output a binary edge image, perform dilation operation to fill small gaps in the edges and eliminate internal holes in the edges, and output a complete contour image; In this embodiment, the binocular camera is calibrated, distortion correction and grayscale conversion are performed on the left image, and then... Gaussian filter Smoothed grayscale image resolution Edge enhancement: Gradient calculation is performed using the Sobel operator, with a horizontal gradient kernel. Vertical gradient kernel gradient magnitude Output edge enhancement map ; Calculate edge enhancement image The global grayscale value is determined using an adaptive thresholding strategy. Extract obstacle edges and output a 2D edge map. At this point, the edge pixel count is 255, and the background pixel count is 0. The dilation operation is performed using... Rectangular structural elements Iterate once to fill in small gaps at the edges and eliminate internal holes at the edges, outputting a complete contour map. .

[0050] S222: Use the eight-neighbor labeling method to label connected regions, traverse the complete contour map output in step S221, label all connected regions, and output the region set; In this embodiment, the eight-neighbor labeling method is used to label connected regions and traverse the contour map. Mark all connected regions and output the set of regions. Each region contains a set of pixel coordinates. ; S223: Output the mask image of the only valid obstacle region after filtering, and extract 3 core feature points by combining the upper surface contour; In this embodiment, the output is the mask image of the unique valid obstacle region after filtering. ; Set of reference points for the front edge of the surface ; Pixel set of reference points at the lower edge of the front surface Based on the segmented Based on the upper surface contour, three core feature points are extracted, including: Midpoint of the front surface edge The obstacle is located at the midpoint of the upper surface on the side closest to the camera. Filter from the set of pixels at the top edge The point at the center of the coordinates Corresponding coordinates .

[0051] Midpoint of the rear surface edge The obstacle is located at the midpoint of the upper surface on the side furthest from the camera. Filter from the set of pixels at the top edge The point with the smallest coordinates (away from the camera) corresponds to the following coordinates. .

[0052] Front surface lower edge reference point Any point on the lower surface of the obstacle closest to the camera, in The lower edge area, filter and point Points with similar coordinates .

[0053] In this embodiment, the SGBM algorithm is used to match only the above three feature points in the right image of the binocular view. The coordinates of the matched points in the right image are as follows: , , .

[0054] S224: Using the SGBM algorithm, only the above 3 feature points are matched in the right image of the binoculars to calculate the obstacle height, obstacle distance and obstacle thickness; In this embodiment, the obstacle height is calculated. Substitute the values ​​into the trigonometric distance measurement formula to measure the distance to the obstacle. ,in for Point parallax, i.e. , For camera Axial focal length, This represents the baseline length of the binocular cameras, i.e., the horizontal distance between the optical centers of the two cameras. The average of 10 consecutive frames is then calculated based on... The criteria are used to remove outliers and obtain the average value. That is, the distance to the obstacle. Extraction and Vertical pixel difference ,have .in For the corresponding camera In this embodiment, the axial focal length is... This offsets minor deviations from the reference point.

[0055] In this embodiment, the barrier thickness is calculated. .calculate Point parallax Substitute into the formula: ,thickness .

[0056] S225: Fuse the information acquired by the ultrasonic sensor with the camera information to correct the obstacle height and obstacle distance obtained in step S224; In this embodiment, information acquired by the ultrasonic sensor is fused with camera information. The ultrasonic sensor is mounted above the front of the left and right suction cups, and its position on the vertical arm is determined, perpendicular to the robot's vertical center. The distance along the axis is The distance relative to the upper end of the vertical arm is Its coordinates in the robot coordinate system are The distance measured by the ultrasonic sensors on both the left and right sides is... and .when and All less than the threshold When an obstacle is detected ahead, the distance to the obstacle is adjusted accordingly. , This is a distance correction factor. On the vertical arm... When moving along the axis, when the original value is less than the threshold value... of or When a mutation occurs, the two vertical arms are at this time The difference in the axial direction is the corrected obstacle height. . Preferably, in this embodiment, we take... , .

[0057] S3 Intelligent Obstacle Avoidance: After obtaining the robot's own posture and position and obstacle feature information in step S2, the robot's movement path is planned and controlled so that the robot can cross the obstacles in front and attach to the next position.

[0058] Step S3 includes accessible scenarios and accessible scenarios. For barrier-free scenarios, it adopts a normal gait, with dual suction cups alternately adsorbing, translating, and rotating, and the rotating motor follows a pre-planned S-shaped curve motion trajectory. For obstacle-prone scenarios, an obstacle-crossing gait is adopted. The suction cup lifting angle, translation distance and adsorption gap are dynamically adjusted through a closed-loop algorithm. Combined with the master-slave motor synchronous control strategy, filtering, feedforward compensation and triple feedback correction are superimposed to suppress jitter. Anomaly adaptive logic is also integrated to adjust the control parameters in real time when the curtain wall shape changes abruptly.

[0059] S31: Normal gait. After the dual-suction cup cleaning robot completes adsorption and establishes a fixed anchor point, it moves forward a certain distance... Greater than And the distance measured by the ultrasonic sensors on both the left and right sides and All greater than At that time, adopt a normal gait.

[0060] During normal gait, in a parallel posture, i.e., the robot... Horizontal bar position And the rotation angle of the two motors At this initial position, with the left vertical arm as support, the suction cup on the left vertical arm maintains adhesion, while the suction cup on the right vertical arm begins to inflate, causing the adhesion force to decrease. The adhesion force decreases until it reaches the vacuum pressure detected by the vacuum sensor and the contact pressure of the pressure sensor. At that time, the right vertical arm completely detaches from the glass curtain wall and is lifted by the motor gear system to... After reaching a certain height, the rate of ascent is... The two vertical arms' bottom rotating motors begin to rotate in the same direction until the two motors rotate at the same angle. Then, the right vertical arm is raised to its current position by the motor gear system. Height. During the lifting process, when the contact pressure of the pressure sensor increases, i.e., the suction cup contacts the glass surface, the suction cup begins to degas and adhere until the vacuum pressure detected by the vacuum sensor is ≥0.09MPa, and the contact pressure detected by the pressure sensor stabilizes in the range of 0.01-0.02MPa. At this point, the suction cup is considered firmly attached. The right vertical arm suction cup remains attached, while the left vertical arm suction cup begins to inflate, reducing the suction force. When the suction force decreases to a level where the vacuum pressure detected by the vacuum sensor and the contact pressure of the pressure sensor are equal, the suction cup is considered firmly attached. At that time, the left vertical arm completely detaches from the glass curtain wall and is lifted by the motor gear system to... After reaching the desired height, the rotary motors at the bottom of the two vertical arms begin to rotate in the same direction until... Then, the left vertical arm is raised to [position] by the motor gear system. Height. During the lifting process, when the contact pressure of the pressure sensor increases, i.e., the suction cup contacts the glass surface, the suction cup begins to expel air for adsorption, until the vacuum pressure detected by the vacuum sensor is ≥0.09MPa, and the contact pressure detected by the pressure sensor stabilizes in the range of 0.01-0.02MPa, at which point the suction cup is considered to be firmly adsorbed. At this point, the robot enters the next parallel posture and begins the next cycle of movement in the reverse order. Preferably, , .

[0061] In the motion of two rotating motors, a pre-planned S-shaped curve trajectory is used. (See also...) Figure 5 The angular velocity of the rotary electric motor follows an S-shaped curve trajectory. The S-shaped curve trajectory represents the rotation of the rotary electric motor to an angle. The complete trajectory cycle consists of an acceleration phase, a deceleration phase, a constant speed phase, an acceleration / deceleration phase, and a deceleration / deceleration phase, in sequence. This represents the rotation to an angle. The variable speed programming process is shown in the table below. Preferably, the maximum angular velocity... Maximum angular acceleration .

[0062]

[0063] In this embodiment, a master-slave architecture and a synchronous control strategy with error correction are adopted in the movement of the dual rotary motors. The rotary motor on the adsorption-stabilized side is designated as the master motor, and the other side as the slave motor. Taking parallel posture movement as an example, the left motor is the master motor, and the right motor is the slave motor. A high-frequency filter is superimposed on the master motor's reference trajectory to avoid jitter caused by high-frequency components in the command itself, and feedforward correction is performed for current fluctuations. The basic trajectory is generated based on an S-curve. A second-order Butterworth low-pass filter is used, with a cutoff frequency of... Used to filter out command noise above the jitter frequency band; filtered command... ,in Angular frequency, The damping coefficient is... It is a complex frequency; and based on the load fluctuations fed back by the current sensor. Correct instructions in advance To counteract the jitter caused by load changes due to robot movement, the feedforward coefficient... The cutoff frequency was determined through actual measurement. Damping coefficient Feedforward coefficient .

[0064] From the actual perspective of motor tracking the main motor Based on one's own practical perspective Control commands for the next moment The planning and instruction correction integrates triple feedback of "synchronization deviation + angular velocity jitter + crossbar rotation jitter" and the correction formula.

[0065] The corrections include synchronization correction and jitter suppression correction. The synchronization correction is used to eliminate long-term synchronization errors of the two motors and suppress low-frequency jitter. This is for synchronization gain. The jitter suppression correction term targets high-frequency instantaneous jitter, correcting it in two dimensions: "angular velocity jitter" and "bar jitter". The real-time angular velocity difference between the two motors is determined by their respective encoders. get, Gain correction for angular velocity The real-time rotational jitter of the horizontal arm is sensed by the robot's posture. Calculations show that... , For negative feedback correction, The sampling period is specified. Meanwhile, considering operational safety, the horizontal bar jitter has a higher priority than static synchronization. Preferred, , , , .

[0066] The lifting action is completed synchronously by the motors on both vertical arm guide rails, and the lifting amount is evenly distributed on both sides. That is, when the lifting distance is... At that time, the crossbar is raised. .

[0067] In this example, both motors also follow an S-shaped motion trajectory. The target angle of motor rotation... The distance the motor travels on the guide rail in one revolution is... Angular velocity and angular acceleration satisfy... , Preferred, Maximum angular velocity Maximum angular acceleration .

[0068] In this example, the same master-slave architecture and real-time error correction synchronization control strategy are used. The basic trajectory is generated based on an S-shaped curve. Filtered instructions , The feedforward coefficient Preferred cutoff frequency Damping coefficient Feedforward coefficient .

[0069] From the actual perspective of motor tracking the main motor The corrected formula is as follows:

[0070] For real-time horizontal shaking of the horizontal arm, sensed by the robot's posture. Calculations show that... , For negative feedback correction, The sampling period is [value]. Preferably, [value]. , , , .

[0071] In this embodiment, when the normal gait does not begin with a parallel posture, when or When the value is not zero, under the condition of stable adsorption on one side, the corresponding vertical arm on the other side is raised sequentially, and rotated to... The position of the adsorption-side motor remains unchanged.

[0072] In this embodiment, when the normal gait does not begin with a parallel posture, the lateral arm is not parallel to the horizontal plane, i.e. The correction is performed using the same movement pattern as the normal gait described above. The correction was performed multiple times, with each correction angle being less than [a certain value]. .

[0073] In this embodiment, when the normal gait does not start with a parallel posture, the horizontal arm is first corrected to be horizontal, and then the motor rotation angle is corrected.

[0074] S32: Obstacle-crossing gait. The closed-loop algorithm is as follows: based on the obstacle parameters perceived by the environment visually and the real-time obstacle perception by ultrasonic waves, after the body rises to the height of the obstacle and the ultrasonic waves no longer detect the obstacle, it rises to a reserved safe height, and the rotation angle is set according to the distance to the obstacle and the thickness of the obstacle to perform the obstacle-crossing rotation action; The master-slave motor synchronization control strategy is as follows: a master-slave architecture is adopted to perform synchronous control of the two motors. The slave motor tracks the actual angle of the master motor and plans the control command for the next moment based on the actual angle, and performs synchronization correction and angular velocity jitter and crossarm rotation jitter suppression correction. The slave motor receives the basic trajectory command and rotates. When receiving the command, a low-pass filter is used to filter out command noise higher than the jitter frequency band, and the load fluctuation compensation command fed back by the current sensor is used to offset the jitter caused by load changes due to robot movement. When the initial posture is not parallel, under the condition of adsorption and stabilization on one side, the corresponding vertical arm on the other side is raised, and the crossarm is rotated to correct the horizontal position. Then, the vertical arms are raised in sequence to correct the motor rotation angle.

[0075] Specifically, when the dual-suction-cup cleaning robot is in a parallel posture, at a distance from the front... for And the distance measured by the ultrasonic sensors on both the left and right sides and All At that time, an obstacle-crossing gait is adopted.

[0076] During obstacle-crossing gait, the distance, height, and thickness of obstacles are obtained through obstacle environment perception. , , Three parameters, at this time there are The movement strategy is the same as the normal gait. At this time, the elevation height... Rotation angle Obstacle avoidance lifting height , For the sensor to sense height, in The ultrasonic sensors on the vertical arm on the moving side can detect obstacles at height, exceeding [a certain height]. At altitude, it can sense obstacles ahead. To maintain a safe distance, the robot is raised to its maximum visual sensing height, and the ultrasonic sensors do not continue to raise it after detecting an obstacle. Height. Preferred. Obstacle avoidance rotation angle .

[0077] In this embodiment, the obstacle height ,thickness At that time, the obstacle crossing condition is met. Preferred option... , =20cm.

[0078] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A smart control method for a dual-suction cup cleaning robot adapted to various glass curtain walls, characterized in that, Includes the following steps: S1 Fixed Anchor Point Establishment: By combining the vacuum degree sensor and pressure sensor on the vacuum suction cup, the dual vacuum suction cups in the dual suction cup cleaning robot are ensured to be in a stable adsorption state, thereby establishing the fixed anchor point; S2 State Awareness: After establishing the fixed anchor point in step S1, the state awareness system is activated to simultaneously complete the perception of the environment and obstacles as well as the perception of its own posture, collect and process multi-source sensor data, and provide accurate input for subsequent control strategy decisions; S3 Intelligent Obstacle Avoidance: After obtaining the robot's own posture and position and obstacle feature information in step S2, the robot's movement path is planned and controlled so that the robot can cross the obstacles in front and attach to the next position.

2. The intelligent control method for a dual-suction cup cleaning robot adapted to multi-form glass curtain walls as described in claim 1, characterized in that, In step S1, the vacuum sensor monitors the vacuum pressure inside the suction cup in real time, and the pressure sensor detects the adhesion pressure between the suction cup's adsorption surface and the wall. When the vacuum pressure detected by the vacuum sensor is ≥0.09MPa, and the adhesion pressure detected by the pressure sensor is stable in the range of 0.01-0.02MPa, it is determined that the suction cup is firmly adsorbed and the fixed anchor point is effectively established.

3. The intelligent control method for a dual-suction cup cleaning robot adapted to multi-form glass curtain walls as described in claim 1, characterized in that, Step S2 specifically includes the following steps: S21 Robot Attitude Perception: Relevant data is acquired through an IMU integrated above the center of the cleaning robot's horizontal arm. After preprocessing the raw data, the robot's roll angle is calculated. Pitch angle Yaw angle The robot's posture calculations are fused and corrected, and the motor rotation angles and speeds recorded by the joint motor encoders are combined to finally output the posture parameters of each part of the robot. S22 Environmental Obstacle Perception: Through binocular cameras and ultrasonic sensors, obstacles in front are calculated to obtain key parameters such as obstacle height, obstacle thickness, and obstacle distance.

4. The intelligent control method for a dual-suction cup cleaning robot adapted to multi-shaped glass curtain walls as described in claim 3, characterized in that, In step S21, the IMU includes a three-axis accelerometer and a three-axis gyroscope, and the relevant data it acquires includes three-axis acceleration. With triaxial angular velocity ; The above data preprocessing method is as follows: A moving average filter is used for preprocessing, specifically, the obtained data including triaxial acceleration is preprocessed... With triaxial angular velocity The sensor data X, with Size of the filter window The sampled data at time t is preprocessed, and the formula for the preprocessed data is: Where: X represents the raw sensor data and ; At the current sampling time, This is the size of the filtering window; This is the filtered data.

5. The intelligent control method for a dual-suction cup cleaning robot adapted to multi-shaped glass curtain walls as described in claim 3, characterized in that, The method for calculating accelerometer and gyroscope data in step S21 is as follows: complementary filtering is used to fuse accelerometer and gyroscope data, the roll angle and pitch angle are roughly calculated using the gravity component of the accelerometer, and then the attitude change is predicted based on the integral of the gyroscope angular velocity.

6. The intelligent control method for a dual-suction cup cleaning robot adapted to multi-form glass curtain walls as described in claim 3, characterized in that, The fusion and correction method in step S21 is as follows: using the yaw angle calibration value obtained by the binocular camera, it is updated every 5 seconds through binocular camera visual feature matching, and the drift is corrected by adaptive fusion weights with the acceleration value; the above adaptive fusion weights are obtained from the vibration intensity, the vibration intensity is evaluated based on the acceleration data, different vibration levels are divided, and the corresponding adaptive fusion weights are obtained according to the vibration level.

7. The intelligent control method for a dual-suction cup cleaning robot adapted to multi-form glass curtain walls as described in claim 3, characterized in that, The specific steps of step S22 are as follows: S221: The binocular camera is calibrated, and distortion correction and grayscale conversion are performed on the left image, followed by a smooth grayscale image after Gaussian filtering; the Sobel operator is used to calculate the gradient and output the edge enhancement image; the global grayscale value of the edge enhancement image is calculated, and an adaptive thresholding strategy is used to extract obstacle edges, output a binary edge image, perform dilation operation to fill small gaps in the edges and eliminate internal holes in the edges, and output a complete contour image; S222: Use the eight-neighbor labeling method to label connected regions, traverse the complete contour map output in step S221, label all connected regions, and output the region set; S223: Output the mask image of the only valid obstacle region after filtering, and extract 3 core feature points by combining the upper surface contour; S224: Using the SGBM algorithm, only the above 3 feature points are matched in the right image of the binoculars to calculate the obstacle height, obstacle distance and obstacle thickness; S225: The information acquired by the ultrasonic sensor is fused with the information from the camera to correct the obstacle height and obstacle distance obtained in step S224.

8. The intelligent control method for a dual-suction cup cleaning robot adapted to multi-form glass curtain walls as described in claim 1, characterized in that, Step S3 includes accessible scenarios and accessible scenarios. For barrier-free scenarios, it adopts a normal gait, with dual suction cups alternately adsorbing, translating, and rotating, and the rotating motor follows a pre-planned S-shaped curve motion trajectory. For obstacle-prone scenarios, an obstacle-crossing gait is adopted. The suction cup lifting angle, translation distance and adsorption gap are dynamically adjusted through a closed-loop algorithm. Combined with the master-slave motor synchronous control strategy, filtering, feedforward compensation and triple feedback correction are superimposed to suppress jitter. Anomaly adaptive logic is also integrated to adjust the control parameters in real time when the curtain wall shape changes abruptly.

9. The intelligent control method for a dual-suction cup cleaning robot adapted to multi-form glass curtain walls as described in claim 8, characterized in that, The S-shaped curve motion trajectory is the rotation of the rotary motor to an angle. The complete trajectory cycle consists of an acceleration phase, a deceleration phase, a constant speed phase, an acceleration / deceleration phase, and a deceleration / deceleration phase.

10. The intelligent control method for a dual-suction cup cleaning robot adapted to multi-form glass curtain walls as described in claim 8, characterized in that, The closed-loop algorithm is as follows: based on the obstacle parameters under environmental visual perception and real-time ultrasonic obstacle perception, after raising to the obstacle height and the ultrasonic waves no longer detect the obstacle, raise to a reserved safe height, set the rotation angle according to the distance to the obstacle and the thickness of the obstacle, and perform obstacle-crossing rotation action. The master-slave motor synchronization control strategy is as follows: a master-slave architecture is adopted to perform synchronous control of the two motors. The slave motor tracks the actual angle of the master motor and plans the control command for the next moment based on the actual angle, and performs synchronization correction and angular velocity jitter and crossarm rotation jitter suppression correction. The slave motor receives the basic trajectory command and rotates. When receiving the command, a low-pass filter is used to filter out command noise higher than the jitter frequency band, and the load fluctuation compensation command fed back by the current sensor is used to offset the jitter caused by load changes due to robot movement. When the initial posture is not parallel, under the condition of adsorption and stabilization on one side, the corresponding vertical arm on the other side is raised, and the crossarm is rotated to correct the horizontal position. Then, the vertical arms are raised in sequence to correct the motor rotation angle.

Images