A pool cleaning robot based on machine vision, control system and method

Abstract

This invention discloses a pool cleaning robot, control system, and method based on machine vision, belonging to the field of robotics technology. It addresses the shortcomings of existing pool cleaning robots in terms of intelligence, environmental adaptability, cleaning effect, and positioning accuracy. The system includes a perception unit, a main control unit, a decision-making unit, a motion unit, an operation unit, a power supply unit, and a human-machine interaction unit. In this invention, the system deeply integrates machine vision technology with dynamic path planning algorithms, changing the traditional fixed-path or random cleaning mode of cleaning robots. The vision-based pool leaf recognition system can distinguish the type and density of stains in real time. Combined with the dynamic path planning algorithm, the robot can autonomously identify highly polluted areas and prioritize their cleaning, achieving precise and personalized management of the cleaning process, thereby improving the robot's intelligence and cleaning efficiency.

Description

Technical Field

[0001] This invention belongs to the field of robotics technology, specifically relating to a pool cleaning robot, control system, and method based on machine vision. Background Technology

[0002] As people's demands for quality of life continue to rise, pool cleaning is receiving increasing attention. Traditional pool cleaning methods mainly rely on manual labor, which is not only labor-intensive and inefficient, but also suffers from incomplete cleaning and unsanitary areas. In recent years, with the development of robotics technology, pool cleaning robots have gradually entered the market, aiming to improve cleaning efficiency, reduce labor costs, and enhance cleaning results. However, existing pool cleaning robots still face many technical bottlenecks and limitations in practical applications, making it difficult to meet the growing demand for high-quality cleaning.

[0003] Traditional pool cleaning robots mostly use fixed or random paths for cleaning, unable to dynamically adjust their cleaning paths based on the actual distribution of dirt in the pool. This lack of targeted cleaning leads to low efficiency and difficulty in ensuring cleaning coverage. For example, heavily soiled areas may not be cleaned thoroughly, while clean areas may be repeatedly cleaned, failing to achieve precise cleaning. Furthermore, pools come in various materials, such as tiles and mosaics, and the underwater environment is complex and variable. Traditional robots are ill-suited to these complex pool types, easily getting stuck or malfunctioning. In addition, traditional cleaning robots struggle to accurately identify stains of different properties, such as leaves, algae, and oil films, and cannot implement differentiated cleaning strategies based on stain type, limiting their application in diverse pool environments. Moreover, existing pool cleaning robots generally use a low-cost positioning solution combining inertial navigation systems (INS) with wheel encoders, which has inherent limitations. On the one hand, the navigation error of INS accumulates and diverges over time, and its drift characteristics cause the positioning error to be proportional to the cube of time, which cannot meet the requirements of long-term operation. On the other hand, the bottom of the pool is smooth, and the robot drive wheels are prone to slipping, which causes the encoder odometer data that depends on the rotation of the wheels to be distorted, thereby introducing huge positioning errors and limiting the application of advanced path planning algorithms.

[0004] Existing pool cleaning robots suffer from significant shortcomings in terms of intelligence, environmental adaptability, cleaning effectiveness, and positioning accuracy, making it difficult to meet the demands of modern pool cleaning for high efficiency, precision, and intelligence. To address these issues, we propose a pool cleaning robot, control system, and method based on machine vision. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by providing a pool cleaning robot, control system, and method based on machine vision. This invention solves the problems that existing pool cleaning robots have significant deficiencies in terms of intelligence, environmental adaptability, cleaning effect, and positioning accuracy, making it difficult to meet the high-efficiency, precise, and intelligent requirements of modern pool cleaning.

[0006] This invention is implemented as follows: This invention provides a machine vision-based water tank cleaning robot control system, which includes:

[0007] The sensing unit is used to collect multimodal data on the robot's internal state and external environment in real time;

[0008] The main control unit is used to acquire multimodal data of the robot's internal state and external environment, preprocess the multimodal data, and output the preprocessed multimodal data.

[0009] The decision-making unit generates motion planning and operation instructions based on the preprocessed multimodal data, and pushes the motion planning and operation instructions to the motion unit and the operation unit;

[0010] The motion unit is used to receive instructions from the decision unit and parse the instructions into driving signals to drive the robot to perform movement, turning and climbing actions.

[0011] The work unit is used to receive instructions from the decision unit, parse the instructions into work signals, and execute the pool cleaning operation;

[0012] The power supply unit is electrically connected to the main control unit, decision-making unit, motion unit, and operation unit, and is used to supply power to the main control unit, decision-making unit, motion unit, and operation unit. The power supply unit includes a lithium battery pack, a charging management module, and a multi-channel voltage conversion circuit.

[0013] The human-machine interaction unit is used to receive user commands and send them to the main control unit. At the same time, it receives and displays the operation status data transmitted back by the robot in real time. The human-machine interaction unit interacts with the main control unit through a wireless communication module.

[0014] Preferably, the sensing unit includes:

[0015] The visual perception module is used to acquire image information of the pool bottom and pool walls;

[0016] The distance sensing module is used to measure the distance between the robot and the pool wall or obstacles in real time;

[0017] The attitude perception module integrates a nine-axis inertial measurement unit (IMU) to detect the robot's own attitude in real time and provide feedback for motion control. The attitude perception module includes a gyroscope sensor, a geomagnetic sensor, and an accelerometer.

[0018] Preferably, the main control unit includes:

[0019] The data preprocessing unit is used to acquire multimodal data of the robot's internal state and external environment, preprocess the multimodal data, and output the preprocessed multimodal data.

[0020] The relative pose estimation unit is used to acquire preprocessed multimodal data, identify multi-directional ultrasonic distance parameters and yaw angles in the multimodal data, estimate the robot's underwater relative pose based on the multi-directional ultrasonic distance parameters and yaw angles, and obtain the pool size parameters and the robot's relative position.

[0021] The anti-slip positioning unit uses a slip positioning method based on the fusion of encoder and inertial sensor to calculate and correct the robot's position, and outputs the corrected robot position.

[0022] Preferably, the method for estimating the robot's underwater relative pose based on multi-directional ultrasonic distance parameters and yaw angle includes:

[0023] Once the pool cleaning robot is powered on and enters the water at any location in the pool, it is driven forward by the motion unit to move close to any pool wall.

[0024] After the pool cleaning robot first encounters the wall, it obtains the distance parameters between the robot and the pool wall or obstacle based on the ultrasonic ranging sensor. According to the distance parameters fed back by the two horizontally distributed ultrasonic ranging sensors, it adjusts the yaw angle of the pool cleaning robot so that the pool cleaning robot is perpendicular to the pool wall. Then the pool cleaning robot rotates 180° in any horizontal direction.

[0025] Record the ultrasonic ranging sensor values ​​D1, D2, and D3 of the water tank cleaning robot in the forward, left, and right directions at this time. D1 is the average of the two forward ultrasonic ranging sensor values. Calculate the length and width of the water tank based on the ultrasonic ranging sensor values ​​D1, D2, and D3 of the water tank cleaning robot. The length of the water tank is D2 + D3, and the width is D1.

[0026] A Cartesian coordinate system is established with one corner of the pool as a reference. The relative position of the pool cleaning robot is calculated based on the length and width of the pool, the yaw angle of the pool cleaning robot, and the values ​​D1, D2, and D3 of the forward, left, and right ultrasonic ranging sensors.

[0027] Preferably, the method for correcting robot position calculation based on the slippage positioning method using encoder and inertial sensor fusion includes:

[0028] The raw data of the three-axis acceleration and three-axis angular velocity of the water tank cleaning robot are collected by the inertial measurement unit (IMU), and the raw data are filtered.

[0029] The absolute heading angle information of the pool cleaning robot is obtained by the geomagnetic sensor. The rotational angular velocity of the drive wheel of the pool cleaning robot is measured by the encoder. Based on the theoretical linear velocity or displacement of the drive wheel radius, when the control command of the drive motor of the pool cleaning robot is detected to be zero and the position time exceeds the preset threshold, it is determined that the pool cleaning robot is in a stationary state and the zero speed correction process is triggered.

[0030] A kinematic model of the robot is established, and a slip detection and adaptive fusion strategy is used to determine whether the pool cleaning robot is slipping. The slip detection and adaptive fusion strategy is as follows: the theoretical linear velocity calculated by the encoder is compared with the actual linear velocity calculated by the INS in real time. It is determined whether the absolute value of the theoretical linear velocity and the actual linear velocity exceeds a preset dynamic threshold. If the absolute value of the theoretical linear velocity and the actual linear velocity exceeds the preset dynamic threshold, the pool cleaning robot is determined to be slipping. If the absolute value of the theoretical linear velocity and the actual linear velocity does not exceed the preset dynamic threshold, the pool cleaning robot is determined to be in a normal state. When the pool cleaning robot is in a normal state, the data obtained by the encoder is used as the main observation source, and the actual linear velocity calculated by the INS is corrected based on the theoretical linear velocity calculated by the encoder. When the pool cleaning robot is determined to be slipping, the weight of the encoder data is reduced or completely removed, and the system is fused with the forward displacement observation of the INS after zero velocity correction and the lateral velocity constraint generated by the nonholonomic constraint, so that the system switches to the dead reckoning mode based on INS.

[0031] Position calculation is achieved through an adaptive weight allocation mechanism: Let the position calculated by INS be P. i The encoder calculates the position as P. e The final position output is:

[0032]

[0033]

[0034] Where α and β represent the first weighting coefficient and the second weighting coefficient; when the velocity difference When the dynamic threshold is exceeded, the second weighting coefficient is automatically adjusted so that the second weighting coefficient β approaches 0. At this point, the final position output is approximately... This ensures that positioning continuity is not affected by encoder distortion during slippage.

[0035] On the other hand, the present invention also provides a machine vision-based pool cleaning robot, the machine vision-based pool cleaning robot comprising:

[0036] The robot body includes the robot's main frame;

[0037] The walking mechanism, installed inside the robot's main frame, is used to assist the robot in moving and operating in the pool.

[0038] The working mechanism, located within the main frame of the robot, is used to perform pool cleaning operations;

[0039] The sensing mechanism is used to collect real-time environmental information of the pool and upload the information to the control mechanism.

[0040] The control mechanism is used to acquire water tank environmental information uploaded by the sensing mechanism in real time, and trigger control commands for the operating mechanism based on the water tank environmental information.

[0041] Preferably, the robot's main frame includes a middle shell component, an upper shell component, a rear shell component, a lower shell component, a left shell component, and a right shell component. The upper shell component and the rear shell component are detachably connected to each other. The middle shell component is detachably connected to the upper shell component. The left shell component and the right shell component are symmetrically arranged on both sides of the middle shell component, and the left shell component and the right shell component are detachably connected to the middle shell component. The lower shell component is detachably connected to the left shell component and the right shell component.

[0042] The walking mechanism includes a track drive motor, a track drive wheel, a track driven wheel, a transmission track, and a cross roller bearing. The track drive motor is fixedly installed inside the robot's main frame. The track drive wheel and the track driven wheel are rotatably installed on the robot's main frame and are rotatably connected to each other through the transmission track. The track driven wheel is rotatably connected to the robot's main frame through the cross roller bearing.

[0043] Preferably, the working mechanism includes a cleaning roller brush, a roller brush drive motor, a synchronous pulley drive wheel, a synchronous pulley driven wheel, a roller brush timing belt, a sludge storage component, and a water pump filtration mechanism. The cleaning roller brush is disposed on the front or lower side of the robot's main frame. The roller brush drive motor is fixedly installed inside the robot's main frame. The output shaft of the roller brush drive motor is fixedly connected to the synchronous pulley drive wheel. The synchronous pulley drive wheel and the synchronous pulley drive wheel are driven by the roller brush timing belt.

[0044] The waste storage component is detachably installed inside the robot's main frame, and a wastewater inlet is provided at the bottom of the waste storage component. A wastewater filter screen is provided on the side wall of the waste storage component. The water pump filtration mechanism includes two sets of symmetrically arranged brushless power water pumps. The brushless power water pumps are used to provide power for the water flow during cleaning, and the brushless power water pumps are connected to the waste storage component.

[0045] Preferably, the sensing mechanism includes a sensing camera, at least one set of ultrasonic ranging sensors, and at least one set of supplementary lights. The ultrasonic ranging sensors are detachably arranged on the outside of the robot body, and the supplementary lights are arranged on both sides of the sensing camera. The sensing camera is used to collect visual images of the pool wall and the pool bottom in real time.

[0046] The control mechanism includes a robot main control board, a storage tube, and a lighting switch. The robot main control board is electrically connected to the power supply unit. The lighting switch is detachably installed in the frame stand. The frame stand is detachably installed in the robot's main frame. The storage tube is installed in the frame stand, and a sealing flange is fixedly installed at the end of the storage tube. At least one set of sealing threaded nuts is installed in the sealing flange.

[0047] A wireless charging component is provided on the rear side of the robot's main frame, and the wireless charging component is electrically connected to the power supply unit.

[0048] On the other hand, the present invention also provides a machine vision-based control method for a pool cleaning robot, which specifically includes:

[0049] The pool environment information is collected by a sensing mechanism. The pool environment information includes image information and distance information. The pool environment information is enhanced and processed by adaptive histogram equalization and noise reduction to obtain the enhanced pool environment information.

[0050] An object detection model integrating traditional vision and deep learning is used to analyze the enhanced pool environment information, calculate the target pixel ratio and distribution uniformity, and quantify the distribution of fallen leaves and sediment; the output includes the target object bounding box coordinates, center point pixel position and confidence score.

[0051] By mapping the target pixel coordinates to the pool world coordinate system using camera calibration parameters and the current underwater relative pose of the pool cleaning robot; combined with the robot's odometry, the real-time relative distance and yaw angle between the pool cleaning robot and the target are obtained.

[0052] Based on the feedback of the angle between the target position and the positive direction of the pool cleaning robot, a proportional-integral-derivative controller is designed to adjust the robot's steering. During the steering process, the target position is continuously updated until the robot is facing the target.

[0053] The robot is controlled to move forward to the target location to perform cleaning, and the cleaning effect is verified in real time. Historical data of the robot's posture during the cleaning process is recorded. After the cleaning is completed, the robot is traced back to the starting point based on the historical status. During the process, the robot continuously identifies the target to be cleaned at the bottom of the pool until no target is detected.

[0054] Compared with the prior art, the embodiments of this application have the following main advantages:

[0055] This invention provides a machine vision-based pool cleaning robot control system. The system deeply integrates machine vision technology with dynamic path planning algorithms, changing the traditional fixed-path or random cleaning mode of cleaning robots. The vision-based pool leaf recognition system can distinguish the type and density of stains in real time. Combined with the dynamic path planning algorithm, the robot can autonomously identify highly polluted areas and prioritize their cleaning, achieving precise and personalized management of the cleaning process. This improves the robot's intelligence and cleaning efficiency. Furthermore, the system addresses the challenges of diverse pool environments (such as different materials like tile or mosaic bottoms) and slippage caused by smooth pool bottoms. It estimates the robot's underwater relative pose based on multi-directional ultrasonic distance parameters and yaw angles, and uses a slippage positioning method combining encoders and inertial sensors to correct the robot's position. This effectively suppresses the accumulation of errors in the inertial navigation system and alleviates positioning distortion caused by wheel slippage, improving the robot's adaptability and positioning accuracy in complex environments.

[0056] In this embodiment of the invention, a method for estimating the underwater relative pose of a robot based on multi-directional ultrasonic distance parameters and yaw angle is provided. This method can measure the size of the pool and automatically plan the cleaning path according to the size of the pool. It also achieves a more accurate underwater relative positioning method at a lower cost, bringing users a more convenient, efficient and low-cost cleaning experience.

[0057] This invention provides a method for correcting robot position calculation based on encoder and inertial sensor fusion for slippage localization. This method integrates data from an inertial navigation system (INS), a geomagnetic sensor, and an encoder, while introducing a zero-velocity correction algorithm and nonholonomic motion constraints to suppress the cubic divergence of inertial navigation errors and solve the wheel slippage problem caused by a smooth pool bottom. It utilizes a geomagnetic sensor to provide precise azimuth angles and removes invalid lateral displacements through kinematic constraints; a zero-velocity correction algorithm is designed to reset error accumulation when the robot intermittently stops; and a velocity difference model between the encoder and INS is constructed to achieve real-time detection of slippage events and adaptive switching of the localization source.

[0058] In this embodiment of the invention, the system adopts a modular system architecture with master-slave collaboration, where functional units interact with each other via a bus. This modular layout not only improves the maintainability and upgrade convenience of the system, but more importantly, it achieves effective fault isolation. An anomaly in a single module / unit is less likely to propagate to the entire system. The master control unit, as the "master" unit, is responsible for high-level decision-making and task scheduling; while each function manager, as the "slave" unit, executes specific control algorithms and real-time tasks. The master-slave collaboration mechanism allows the system to perform efficient task allocation and resource scheduling.

[0059] In this embodiment of the invention, the machine vision-based pool cleaning robot integrates both brushing and water pump filtration functions, constructing a complete closed-loop operation of brushing-collection-filtration. The efficient operating mechanism can peel off the dirt attached to the pool surface, while the symmetrically distributed brushless power water pump and the optimized filtration system can collect suspended dirt to the dirt storage component in a timely manner, effectively preventing secondary pollution caused by the re-settling of the dirt washed off. Attached Figure Description

[0060] Figure 1 A schematic diagram of the control system architecture for a pool cleaning robot based on machine vision is shown.

[0061] Figure 2 A schematic diagram of the implementation process of a method for estimating the underwater relative pose of a robot based on multi-directional ultrasonic distance parameters and yaw angle is shown.

[0062] Figure 3 A schematic diagram of robot positioning with a yaw angle of 0° is shown.

[0063] Figure 4 A schematic diagram of robot positioning with a yaw angle of 90° is shown.

[0064] Figure 5 A schematic diagram of robot positioning with a yaw angle of 180° is shown.

[0065] Figure 6 A schematic diagram of robot positioning with a yaw angle of 270° is shown.

[0066] Figure 7 A schematic diagram of the implementation process of the robot position calculation and correction method based on the slip positioning method of encoder and inertial sensor fusion is shown.

[0067] Figure 8 A schematic diagram illustrating the principle of the slip detection and adaptive fusion strategy is shown.

[0068] Figure 9 A schematic diagram illustrating the implementation process of a machine vision-based control method for a water tank cleaning robot is shown.

[0069] Figure 10A schematic diagram of the overall structure of the water tank cleaning robot in an embodiment of the present invention is shown.

[0070] Figure 11 A cross-sectional view of a water tank cleaning robot according to an embodiment of the present invention is shown.

[0071] Figure 12 A schematic diagram of the robot's main frame structure is shown in an embodiment of the present invention.

[0072] Figure 13 A schematic diagram of the internal structure of the pool cleaning robot in an embodiment of the present invention is shown.

[0073] Figure 14 A schematic diagram of the walking mechanism in an embodiment of the present invention is shown.

[0074] Figure 15 A schematic diagram of the working mechanism in an embodiment of the present invention is shown.

[0075] Figure 16 A schematic diagram of the sensing mechanism in an embodiment of the present invention is shown.

[0076] Figure 17 A schematic diagram of the control mechanism in an embodiment of the present invention is shown.

[0077] In the diagram: 1-Robot body, 2-Robot main frame, 3-Walking mechanism, 4-Operating mechanism, 5-Sensing mechanism, 6-Wireless charging component, 7-Control mechanism, 8-Sealing wire nut, 9-Sealing flange, 10-Warehouse tube, 11-Track drive motor, 12-Track drive wheel, 13-Track driven wheel, 14-Cross roller bearing, 15-Transmission track, 16-Right side shell component, 17-Roller brush drive motor, 18-Synchronous pulley drive wheel, 19-Roller brush synchronous belt, 20-Synchronous pulley driven wheel, 21-Cleaning roller brush, 22-Water pump. Filter mechanism, 23-Lighting switch, 24-Frame stand, 25-Sewage storage inlet, 26-Supplemental light, 27-Sensing camera, 28-Ultrasonic ranging sensor, 29-Sewage storage component, 30-Sewage filter screen, 31-Middle shell component, 32-Lower shell component, 33-Left shell component, 34-Rear shell component, 35-Upper shell component, 100-Sensing unit, 200-Main control unit, 300-Decision unit, 400-Motion unit, 500-Operating unit, 600-Power supply unit, 700-Human-machine interaction unit. Detailed Implementation

[0078] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0079] Existing pool cleaning robots have significant shortcomings in terms of intelligence, environmental adaptability, cleaning effect, and positioning accuracy, making it difficult to meet the high-efficiency, precise, and intelligent requirements of modern pool cleaning. To address these issues, we propose a pool cleaning robot, control system, and method based on machine vision. In short, the system consists of a sensing unit 100, a main control unit 200, a decision-making unit 300, a motion unit 400, an operation unit 500, a power supply unit 600, and a human-machine interaction unit 700. The system architecture realizes closed-loop automatic control from environmental information acquisition and intelligent decision-making to motion execution. In this embodiment, a pool cleaning robot control system based on machine vision is provided. The system changes the working mode of traditional cleaning robots that use fixed paths or random cleaning by deeply integrating machine vision technology with dynamic path planning algorithms. A vision-based water tank leaf recognition system can distinguish the type and density of stains in real time. Combined with a dynamic path planning algorithm, the robot can autonomously identify highly polluted areas and prioritize their cleaning, achieving precise and personalized management of the cleaning process. This improves the robot's intelligence and cleaning efficiency. In addition, the system can address the challenges of diverse water tank environments (such as different materials like tiles and mosaics on the bottom) and slippage issues caused by smooth bottoms. It estimates the robot's underwater relative pose based on multi-directional ultrasonic distance parameters and yaw angles, and uses a slippage positioning method that combines encoders and inertial sensors to correct the robot's position. This effectively suppresses the accumulation of errors in the inertial navigation system and alleviates positioning distortion caused by wheel slippage, thus improving the robot's adaptability and positioning accuracy in complex environments.

[0080] This invention provides a machine vision-based control system for a water tank cleaning robot. Figure 1 A schematic diagram of the control system architecture for a machine vision-based pool cleaning robot is shown. The machine vision-based pool cleaning robot control system specifically includes:

[0081] The sensing unit 100 is used to collect multimodal data of the robot's internal state and external environment in real time.

[0082] In this embodiment of the invention, the sensing unit 100 includes:

[0083] The visual perception module, which is used to acquire image information of the pool bottom and pool wall, consists of a perception camera 27 and a supplementary light 26, and is used to acquire image information of the pool bottom and pool wall to identify stains, obstacles, etc.

[0084] The distance sensing module is used to measure the distance between the robot and the pool wall or obstacles in real time. It consists of multiple ultrasonic ranging sensors 28 arranged on the front, left and right sides of the robot. In particular, the forward dual sensors can help determine whether the robot's posture is perpendicular to the wall by measuring the difference in the measurement values.

[0085] The attitude perception module, integrating a nine-axis inertial measurement unit (IMU), is used to detect the robot's own attitude in real time and provide feedback for motion control. This module includes a gyroscope sensor, a magnetometer sensor, and an accelerometer. However, the nine-axis IMU itself has inherent errors, such as zero bias in the accelerometer, zero bias drift in the gyroscope, and the soft and hard iron effects of the magnetometer sensor. To obtain high-precision attitude data, sensor data fusion is required. Algorithms such as extended Kalman filters or unscented Kalman filters are used to fuse complementary information from the accelerometer, gyroscope, and magnetometer sensor (accelerometers provide a gravity direction reference but contain motion noise, gyroscopes provide short-term accurate angular velocities but are subject to drift, and magnetometer sensors provide absolute heading references but are susceptible to interference) to estimate more accurate attitude angles (pitch, roll, and yaw).

[0086] The main control unit 200 is used to acquire multimodal data of the robot's internal state and external environment, preprocess the multimodal data, and output the preprocessed multimodal data. It should be noted that the main control unit 200 serves as the core processing and decision-making unit 300 of the system. It can receive raw image data collected by the sensing camera 27 and data such as three-axis acceleration and angular velocity collected by the gyroscope sensor, and process the acquired data. The main control unit 200 interacts with external units via a communication interface (serial communication). The main control unit 200 is an embedded computing platform with multi-source data fusion processing, task scheduling, and system coordination functions. The controller also has corresponding data processing capabilities, such as image preprocessing and feature extraction, inertial data filtering and attitude calculation, and fusion and calibration of multi-channel distance information. Furthermore, the main control unit 200 can generate motion planning and operation instructions in real time based on environmental information and built-in algorithms, and issue high-level instructions (such as target heading, target speed, and cleaning start / stop signals) to each execution unit. The main control unit 200 also has multi-channel communication management capabilities to ensure stable, reliable, and real-time transmission of the internal command stream, thereby achieving unified coordination and scheduling of the work of each unit, meeting the timing requirements of real-time control tasks, and adopting necessary hardware and software fault tolerance mechanisms to ensure the basic operational capability of the system under abnormal sensor data or external interference.

[0087] The main control unit 200 includes:

[0088] The data preprocessing unit is used to acquire multimodal data of the robot's internal state and external environment, preprocess the multimodal data, and output the preprocessed multimodal data.

[0089] The relative pose estimation unit is used to acquire preprocessed multimodal data, identify multi-directional ultrasonic distance parameters and yaw angles in the multimodal data, estimate the robot's underwater relative pose based on the multi-directional ultrasonic distance parameters and yaw angles, and obtain the pool size parameters and the robot's relative position.

[0090] The anti-slip positioning unit uses a slip positioning method based on the fusion of encoder and inertial sensor to calculate and correct the robot's position, and outputs the corrected robot position.

[0091] The decision-making unit 300 generates motion planning and operation instructions based on preprocessed multimodal data, and pushes these instructions to the motion unit 400 and the operation unit 500. As the intelligent hub of the system, the decision-making unit 300 receives and processes multi-source information from the sensing unit 100. Its built-in control strategies and algorithms fuse and analyze the data to generate high-level motion and operation instructions, which are then distributed to the motion unit 400 and the operation unit 500.

[0092] The motion unit 400 receives instructions from the decision unit 300 and parses them into drive signals to drive the robot to perform movement, turning, and wall-climbing actions. The motion unit 400 can receive instructions from the decision unit 300. The motion unit 400 can control the robot's track drive motor 11. Its main function is to parse and convert high-level instructions into specific drive signals to precisely control the track drive motor 11 that drives the tracks, enabling the robot to move, turn, and climb. The track drive motor 11 has precise speed and torque control capabilities, meaning the motor needs to maintain a stable speed output under different loads (such as the gravitational component when traveling on flat ground or climbing a wall) to ensure the robot moves at a predetermined speed.

[0093] The work unit 500 receives instructions from the decision unit 300, parses these instructions into work signals, and executes the pool cleaning operation. As the system's execution layer, the work unit 500 receives instructions from the decision unit 300. It includes the robot's cleaning roller brush 21, the roller brush drive motor 17 that drives the cleaning roller brush 21, and the water pump motor, among other cleaning mechanisms. Its main function is to parse and convert high-level instructions into specific drive signals to precisely control the start / stop and rotation speed of the roller brush, as well as the water pump flow rate, thereby efficiently completing the cleaning task.

[0094] The roller brush drive motor 17 features low-speed, high-torque output characteristics and strong overload capacity to accommodate resistance fluctuations that the roller brush may encounter during cleaning, preventing stalling and shutdown due to dirt entanglement or sudden load increases. All motors in the working unit 500 and motion unit 400 must have an enclosure protection rating of at least IP68 (according to GB / T4208 standard). This rating requires the motor to withstand the effects of continuous submersion under manufacturer-specified conditions (typically hydrostatic pressure environment with a water depth of at least 3 meters), ensuring that the amount of water entering the motor does not reach harmful levels.

[0095] The power supply unit 600 is electrically connected to the main control unit 200, decision-making unit 300, motion unit 400, and operation unit 500. It is used to supply power to the main control unit 200, decision-making unit 300, motion unit 400, and operation unit 500. The power supply unit 600 includes a lithium battery pack, a charging management module, and a multi-channel voltage conversion circuit. The power supply unit 600 can convert and distribute the battery voltage to different stable levels (such as 12V, 8.4V, and 5V) to provide isolated and stable power support for the main control unit 200, motor, sensor, and lighting system in the system. The power supply unit 600 has complete protection functions, which may be characterized by the integration of overcurrent protection, overvoltage protection, undervoltage lockout, and overtemperature protection circuits. The charging management module needs to support intelligent management of the entire cycle of trickle charging, constant current charging, constant voltage charging, and charging termination, and has battery reverse connection protection and thermal regulation functions to ensure the safety and lifespan of the lithium battery pack.

[0096] The human-machine interaction unit 700 is used to receive user instructions and send them to the main control unit 200. At the same time, it receives and displays the operation status data transmitted back by the robot in real time. The human-machine interaction unit 700 interacts with the main control unit 200 through a wireless communication module.

[0097] It should be noted that the graphical user interface adopts a database-based configuration method, mapping and associating interface controls (such as function buttons) with corresponding device control commands or data display pages. When adjustments or expansions in functionality are needed, this can be achieved by updating the database configuration without modifying the program source code, thereby improving the system's maintainability and flexibility. The human-machine interface unit 700, as the system's remote monitoring and control terminal, has the capability to wirelessly communicate with the main control unit 200. This unit establishes a stable wireless data link with the main control unit 200 by integrating a wireless communication module (a Wi-Fi module or Bluetooth module supporting the IEEE 802.11 protocol). Its communication protocol is based on the TCP / IP stack and can utilize lightweight application protocols such as MQTT to achieve efficient and reliable transmission of control commands and status information.

[0098] Specifically, the human-machine interface unit 700 sends task instructions, parameter settings, and other control signals to the main control unit 200 via this wireless link, and receives robot posture, work progress, sensor data, and alarm information from the main control unit 200 in real time. This wireless communication design allows users to remotely and non-contactly monitor and intervene in the pool cleaning robot, enhancing the convenience and flexibility of the system and providing key support for building an intelligent pool operation and maintenance system.

[0099] This invention provides a machine vision-based pool cleaning robot control system. The system deeply integrates machine vision technology with dynamic path planning algorithms, changing the traditional fixed-path or random cleaning mode of cleaning robots. The vision-based pool leaf recognition system can distinguish the type and density of stains in real time. Combined with the dynamic path planning algorithm, the robot can autonomously identify highly polluted areas and prioritize their cleaning, achieving precise and personalized management of the cleaning process. This improves the robot's intelligence and cleaning efficiency. Furthermore, the system addresses the challenges of diverse pool environments (such as different materials like tile or mosaic bottoms) and slippage caused by smooth pool bottoms. It estimates the robot's underwater relative pose based on multi-directional ultrasonic distance parameters and yaw angles, and uses a slippage positioning method combining encoders and inertial sensors to correct the robot's position. This effectively suppresses the accumulation of errors in the inertial navigation system and alleviates positioning distortion caused by wheel slippage, improving the robot's adaptability and positioning accuracy in complex environments.

[0100] In a further preferred embodiment of the present invention, a method for estimating the underwater relative pose of a robot based on multi-directional ultrasonic distance parameters and yaw angle is also provided. Figure 2 This diagram illustrates the implementation flow of a method for estimating the underwater relative pose of a robot based on multi-directional ultrasonic range parameters and yaw angle. The method specifically includes:

[0101] S101, the pool cleaning robot is powered on and enters the water at any position in the pool. The motion unit 400 drives the pool cleaning robot to walk forward to approach any pool wall.

[0102] S102, after the pool cleaning robot hits the wall for the first time, it obtains the distance parameters between the robot and the pool wall or obstacle based on the ultrasonic ranging sensor 28. According to the distance parameters fed back by the two horizontally distributed ultrasonic ranging sensors 28, it adjusts the yaw angle of the pool cleaning robot so that the pool cleaning robot is perpendicular to the pool wall. Then the pool cleaning robot rotates 180° in any horizontal direction.

[0103] S103, record the values ​​D1, D2, and D3 of the ultrasonic ranging sensors 28 of the pool cleaning robot in the forward, left, and right directions at this time. D1 is the average of the values ​​of the two forward ultrasonic ranging sensors 28. Calculate the length and width of the pool based on the values ​​D1, D2, and D3 of the ultrasonic ranging sensors 28 of the pool cleaning robot. The length of the pool is D2 + D3, and the width is D1.

[0104] S104. Establish a Cartesian coordinate system with one corner of the pool as the reference. Calculate the relative position of the pool cleaning robot based on the length and width of the pool, the yaw angle of the pool cleaning robot, and the values ​​D1, D2, and D3 of the forward, left, and right ultrasonic ranging sensors 28.

[0105] In this embodiment of the invention, a method for estimating the underwater relative pose of a robot based on multi-directional ultrasonic distance parameters and yaw angle is provided. This method can measure the size of the pool and automatically plan the cleaning path according to the size of the pool. It also achieves a more accurate underwater relative positioning method at a lower cost, bringing users a more convenient, efficient and low-cost cleaning experience.

[0106] In this embodiment, the method for calculating the relative position of the pool cleaning robot, based on the yaw angle and the forward, leftward, and rightward ultrasonic ranging sensor values ​​D1, D2, and D3 of the ultrasonic ranging sensor 28, includes:

[0107] Taking into account the robot's characteristic of moving only in directions parallel or perpendicular to the wall, i.e., yaw angle The angles can only be 0°, 90°, 180°, or 270°. The length L = D1 + D3 and width W = D1 of the pool have been measured in step S103. Establish a coordinate system: with the pool wall initially approached by the robot as the x-axis, its vertical direction as the y-axis, and the lower left corner of the pool as the origin O(0, 0). Therefore, the range of the pool is 0 ≤ x ≤ L, 0 ≤ y ≤ W.

[0108] Based on the current yaw angle Given the real-time ultrasonic ranging values ​​D1 (average forward distance), D2 (leftward distance), and D3 (rightward distance), the formula for calculating the robot's coordinates (x, y) is as follows:

[0109] like Figure 3 As shown, a schematic diagram of robot positioning with a yaw angle of 0° is presented. (The robot's forward direction is parallel to the positive x-axis):

[0110] The forward sensor points to the positive x-axis and measures the distance to the right side wall (x=L): D1=Lx→x=L-D1;

[0111] The sensor pointing to the left points in the positive y-axis direction and measures the distance to the upper sidewall (y=W): D2=Wy→y=W-D2;

[0112] The right-hand sensor points to the negative y-axis and measures the distance to the lower sidewall (y=0): D3=y;

[0113] From D2+D3=W, we get y=D3 or y=W-D2, which are equivalent. We can directly take: x=L-D1, y=D3;

[0114] like Figure 4 As shown, a schematic diagram of robot positioning with a yaw angle of 90° is presented. (The robot's forward direction is parallel to the positive y-axis, that is, perpendicular to the initial wall surface towards the inside of the pool):

[0115] The forward sensor points to the positive y-axis and measures the distance to the upper sidewall (y=W): D1=Wy→y=W-D1;

[0116] The left-hand sensor points to the negative x-axis and measures the distance to the left side wall (x=0): D2=x;

[0117] The right-hand sensor points to the positive x-axis and measures the distance to the right side wall (x=L): D3=Lx. From D2+D3=L, we get x=D2 or x=L-D3. We can directly take: x=D2, y=W-D1.

[0118] like Figure 5 As shown, a schematic diagram of robot positioning with a yaw angle of 180° is illustrated. (The robot's forward direction is parallel to the negative x-axis):

[0119] The forward sensor points to the negative x-axis and measures the distance to the left side wall (x=0): D1=x→x=D1;

[0120] The left-hand sensor points to the negative y-axis and measures the distance to the lower sidewall (y=0): D2=y;

[0121] The right-hand sensor points to the positive y-axis and measures the distance to the upper sidewall (y=W): D3=Wy. From D2+D3=W, we get y=D2 or y=W-D3. We can directly take: x=D1, y=D2.

[0122] like Figure 6 As shown, a schematic diagram of robot positioning at a yaw angle of 270° is illustrated. (The robot's forward direction is parallel to the negative y-axis, that is, perpendicular to the initial wall surface):

[0123] The forward sensor points to the negative y-axis and measures the distance to the lower sidewall (y=0): D1=y→y=D1;

[0124] The sensor pointing to the left is aligned with the positive x-axis and measures the distance to the right side wall (x=L): D2=Lx→x=L-D2;

[0125] The right-hand sensor points to the negative x-axis and measures the distance to the left side wall (x=0): D3=x;

[0126] From D2+D3=L, we get x=D3 or x=L-D2, so we can directly take: x=D3, y=D1.

[0127] In summary, the robot's relative position can be determined based on the current yaw angle. The solution is obtained by substituting the corresponding formulas. This method utilizes the simplification of trigonometric function values ​​under discrete yaw angles (sin0°=0, sin90°=1, sin180°=0, sin270°=-1; cosine is similar), avoiding trigonometric function calculations under continuous angles, reducing computational complexity, and conforming to the constraint that the robot only moves parallel or perpendicular to the wall, thus achieving rapid relative positioning at low cost.

[0128] In a further preferred embodiment of the present invention, a method for correcting robot position calculation based on a slippage positioning method using encoder and inertial sensor fusion is also provided. Figure 7 A schematic diagram illustrating the implementation process of a robot position calculation correction method based on encoder and inertial sensor fusion for slip positioning is shown. The method specifically includes:

[0129] S201: Based on the inertial measurement unit (IMU), the raw data of the three-axis acceleration and three-axis angular velocity of the water tank cleaning robot are collected and the raw data are filtered.

[0130] In this embodiment of the invention, the robot's three-axis acceleration and three-axis angular velocity are continuously collected by an onboard inertial measurement unit (IMU), and the raw data is filtered to suppress high-frequency noise and random errors. The robot's absolute heading angle information is obtained by a geomagnetic sensor to compensate for the heading drift of the gyroscope. The rotational angular velocity of the drive wheels is measured by an encoder and converted into theoretical linear velocity or displacement based on the wheel radius. When the robot's drive motor control command is detected to be zero and the duration exceeds a preset threshold, the robot is determined to be in a stationary state, triggering a zero-speed correction process. At this time, the velocity information calculated by the INS is used as an observation to correct the navigation parameters.

[0131] S202: Based on the geomagnetic sensor, the absolute heading angle information of the water tank cleaning robot is obtained. The rotational angular velocity of the drive wheel of the water tank cleaning robot is measured by the encoder. The motor command status uploaded by the motor controller is obtained for zero speed detection to determine whether the robot is stationary. Based on the theoretical linear velocity or displacement of the drive wheel radius, when the drive motor control command of the water tank cleaning robot is detected to be zero output and the position time exceeds the preset threshold, it is determined that the water tank cleaning robot is stationary and the zero speed correction process is triggered.

[0132] S203, establish the kinematic model of the robot, and determine whether the pool cleaning robot is slipping based on a slip detection and adaptive fusion strategy. Figure 8The diagram illustrates the principle of the slip detection and adaptive fusion strategy, which involves real-time comparison of the theoretical linear velocity calculated by the encoder with the actual linear velocity solved by the INS.

[0133] Here, INS represents the inertial navigation system, with a theoretical linear velocity. , These are the encoder angular velocity and the wheel radius, respectively.

[0134] The system determines whether the absolute value of the theoretical linear velocity and the actual linear velocity exceeds a preset dynamic threshold. If the absolute value exceeds the preset dynamic threshold, the water cleaning robot is determined to be in a slipping state. If the absolute value does not exceed the preset dynamic threshold, the water cleaning robot is determined to be in a normal state. Based on the nonholonomic constraint characteristics of the robot using a tracked or differential drive structure, its motion is modeled as being able to move only in the forward / backward direction, with a theoretical lateral velocity of zero. In the carrier coordinate system defined by the heading angle provided by the magnetometer, all non-zero lateral velocities calculated by the INS are considered error observations and are discarded. When the water cleaning robot is in a normal state (normal mode), the data obtained by the encoder is used as the main observation source, and the actual linear velocity calculated by the INS is corrected based on the theoretical linear velocity calculated by the encoder. When the water cleaning robot is determined to be in a slipping state (slipping mode), the weight of the encoder data is reduced or it is completely discarded. Instead, the system relies on the INS forward displacement observation after zero-velocity correction and the lateral velocity constraints generated by the nonholonomic constraints for fusion, so that the system switches to a dead reckoning mode dominated by INS.

[0135] S204, Position calculation is achieved through an adaptive weight allocation mechanism: Let the position calculated by INS be P. i The encoder calculates the position as P. e The final position output is:

[0136]

[0137]

[0138] Where α and β represent the first weighting coefficient and the second weighting coefficient; when the velocity difference When the dynamic threshold is exceeded, the second weighting coefficient is automatically adjusted so that the second weighting coefficient β approaches 0. At this point, the final position output is approximately... This ensures that positioning continuity is not affected by encoder distortion during slippage.

[0139] This invention provides a method for correcting robot position calculation based on slippage localization using encoder and inertial sensor fusion. This method integrates data from an inertial navigation system (INS), a geomagnetic sensor, and an encoder, while introducing a zero-velocity correction algorithm and nonholonomic motion constraints to suppress cubic divergence of inertial navigation errors and address wheel slippage issues caused by smooth pool bottoms. It utilizes a geomagnetic sensor to provide precise azimuth angles and removes invalid lateral displacements through kinematic constraints; a zero-velocity correction algorithm is designed to reset error accumulation when the robot intermittently stops; and a velocity difference model between the encoder and INS is constructed to achieve real-time detection of slippage events and adaptive switching of the localization source.

[0140] In this embodiment of the invention, the system adopts a modular system architecture with master-slave collaboration, where functional units interact via a bus. This modular layout not only improves the maintainability and upgrade convenience of the system, but more importantly, it achieves effective fault isolation. Anomalies in a single module / unit are less likely to propagate to the entire system. The master control unit 200, acting as the "master" unit, is responsible for high-level decision-making and task scheduling; while each function manager, acting as a "slave" unit, executes specific control algorithms and real-time tasks. The master-slave collaboration mechanism allows for efficient task allocation and resource scheduling, while effectively controlling manufacturing costs while ensuring high performance. The relative pose estimation method simplifies computational complexity by utilizing the structural constraints of a square pool, avoiding expensive underwater positioning equipment; and the slip positioning method based on encoder and inertial sensor fusion compensates for hardware limitations through software algorithms, significantly improving the utilization efficiency of existing sensors.

[0141] This invention also provides a machine vision-based control method for a pool cleaning robot. This method is suitable for applications where, after the initial cleaning, precise cleaning of the remaining small amount of fallen leaves and other debris in the pool is required, or for situations where only a small amount of fallen leaves remain in the pool. Figure 9 This diagram illustrates the implementation flow of a machine vision-based control method for a water tank cleaning robot. The specific steps of this machine vision-based control method include:

[0142] S10, Image Acquisition and Enhancement Processing: The pool environment information is acquired through the sensing mechanism 5. The pool environment information includes image information and distance information. The pool environment information is enhanced and processed by adaptive histogram equalization and noise reduction to obtain the enhanced pool environment information.

[0143] S20, Leaf and Sediment Recognition and Analysis: This section employs a target detection model that integrates traditional vision and deep learning to analyze the enhanced pool environment information, calculating the target pixel ratio and distribution uniformity to quantify the distribution of fallen leaves and sediment. The output includes the target object's bounding box coordinates, center point pixel position, and confidence level. When multiple objects to be cleaned are detected during the leaf and sediment recognition analysis, for each detected target, its appearance and motion features are extracted and matched with existing trajectories. A unique identifier (ID) and corresponding motion trajectory are assigned and maintained for each target. Based on each target's initial detection confidence, trajectory stability, and appearance feature consistency, its comprehensive confidence level is calculated. The comprehensive confidence levels of all tracked targets are compared, and the target with the highest comprehensive confidence level is identified as the primary cleaning target, even if its comprehensive confidence level drops below the highest during execution. In the comprehensive confidence level evaluation step, trajectory stability is evaluated based on the number of frames the target has been continuously tracked, and appearance feature consistency is evaluated based on the cosine similarity of the target's appearance feature vectors during tracking.

[0144] S30, Target Position Mapping and Robot Localization: By using camera calibration parameters and the current underwater relative pose of the pool cleaning robot, the target pixel coordinates are mapped to the pool world coordinate system; combined with the robot odometry, the relative distance and yaw angle between the pool cleaning robot and the target are measured in real time. If there are multiple targets, the cleaning order is dynamically sorted according to the nearest priority principle or the cluster analysis results.

[0145] S40, Dynamic Steering and Approach Control: Based on feedback from the target position and the positive angle between the target position and the water cleaning robot, a proportional-integral-derivative controller is designed to adjust the robot's steering. During the steering process, the target position is continuously updated until the robot is directly facing the target (the angle is less than a set threshold). Visual servo control ensures that the robot remains aligned with the target while moving forward.

[0146] S50, Cleaning Operation and Status Retrospective: Controls the pool cleaning robot to move forward to the target position to perform cleaning (such as suction or brushing), and verifies the cleaning effect in real time (such as through secondary image analysis), records the robot's pose history data during the cleaning process, and retrospectively returns to the cleaning starting point based on the historical status after cleaning. During the process, it continuously identifies the target to be cleaned at the bottom of the pool until no target is detected.

[0147] In this embodiment, the image enhancement processing includes color correction to compensate for the absorption of red light by the water, contrast enhancement to enhance the difference between the target and the background, and suspended particle noise filtering. The denoising processing adopts an underwater image denoising model based on deep learning. The cleaning object also includes: if a new target or an existing target with a higher overall confidence appears during the movement toward the primary cleaning object, the cleaning object is updated in real time according to the new priority determination result. However, after the movement ends, the target is not locked based on the highest overall confidence.

[0148] This invention also provides a pool cleaning robot based on machine vision, such as... Figures 10-13 As shown, the machine vision-based pool cleaning robot specifically includes:

[0149] Robot body 1, which includes robot main frame 2;

[0150] The walking mechanism 3 is installed inside the robot's main frame 2 to assist the robot body 1 in moving and operating in the pool.

[0151] The working mechanism 4 is set inside the robot's main frame 2 and is used to perform pool cleaning operations;

[0152] Sensing mechanism 5 is used to collect water pool environmental information in real time and upload the water pool environmental information to control mechanism 7;

[0153] The control mechanism 7 is used to acquire the water pool environment information uploaded by the sensing mechanism 5 in real time, and trigger the control command of the operation mechanism 4 based on the water pool environment information.

[0154] In this embodiment of the invention, the machine vision-based pool cleaning robot realizes the automation and precision of cleaning operations by sensing the pool environment information in real time and intelligently triggering the control commands of the working mechanism 4. This significantly improves cleaning efficiency and quality, reduces the cost of manual intervention, and enhances the adaptability to complex pool environments. It effectively avoids problems such as insufficient coverage, incomplete cleaning, and secondary pollution that exist in traditional cleaning methods, providing an efficient, intelligent, and reliable solution for the field of pool cleaning.

[0155] In this embodiment of the invention, the robot's main frame 2 includes a middle shell component 31, an upper shell component 35, a rear shell component 34, a lower shell component 32, a left shell component 33, and a right shell component 16. The upper shell component 35 and the rear shell component 34 are detachably connected. The middle shell component 31 is detachably connected to the upper shell component 35. The left shell component 33 and the right shell component 16 are symmetrically arranged on both sides of the middle shell component 31, and the left shell component 33 and the right shell component 16 are detachably connected to the middle shell component 31. The lower shell component 32 is detachably connected to the left shell component 33 and the right shell component 16, respectively. This design balances lightweight, high protection, and structural strength, providing a fundamental guarantee for the high performance and reliability of the pool cleaning robot.

[0156] like Figure 14 As shown, the walking mechanism 3 includes a track drive motor 11, a track drive wheel 12, a track driven wheel 13, a transmission track 15, and a cross roller bearing 14. The track drive motor 11 is fixedly installed inside the robot's main frame 2. The track drive wheel 12 and the track driven wheel 13 are rotatably installed on the robot's main frame 2, and the track drive wheel 12 and the track driven wheel 13 are rotatably connected to each other through the transmission track 15. The track driven wheel 13 is rotatably connected to the robot's main frame 2 through the cross roller bearing 14. The walking mechanism 3 independently drives the tracks on both sides through two track drive motors 11. By coordinating the speed and direction of the two motors, the robot can move forward, backward, and in a straight line.

[0157] In a further preferred embodiment of the present invention, such as Figure 15 As shown, the working mechanism 4 includes a cleaning roller brush 21, a roller brush drive motor 17, a synchronous pulley drive wheel 18, a synchronous pulley driven wheel 20, a roller brush timing belt 19, a dirt storage component 29, and a water pump filtration mechanism 22. The cleaning roller brush 21 is located on the front or lower side of the robot's main frame 2. The roller brush drive motor 17 is fixedly installed inside the robot's main frame 2. The output shaft of the roller brush drive motor 17 is fixedly connected to the synchronous pulley drive wheel 18. The synchronous pulley drive wheel 18 and the synchronous pulley driven wheel 20 are meshed and driven by the roller brush timing belt 19, ensuring that the rotation speed of the cleaning roller brush 21 is strictly synchronized with the output speed of the roller brush drive motor 17, thus ensuring the stability and consistency of the roller brush rotation speed during cleaning operations. The roller brush drive motor 17 is bolted to the middle outer shell component 31. The function of the cleaning roller brush 21 is to clean the bottom, side walls, or water level of the pool during rotation.

[0158] In this embodiment of the invention, the machine vision-based pool cleaning robot integrates both brushing and water pump filtration functions, constructing a complete closed loop of brushing-collection-filtration operations. The efficient operating mechanism 4 can peel off the dirt attached to the pool surface, while the symmetrically distributed brushless power water pump and the optimized filtration system can promptly collect suspended dirt into the dirt storage component 29, effectively preventing secondary pollution caused by the re-settling of the brushed dirt.

[0159] The waste storage component 29 is detachably installed within the robot's main frame 2, allowing for easy removal and placement for convenient daily cleaning and maintenance. The bottom of the waste storage component 29 has a wastewater inlet 25, and its side walls are equipped with wastewater filters 30. The water pump filtration mechanism 22 includes two symmetrically arranged brushless water pumps. These pumps provide power for the water flow during cleaning. The pumps are connected to the waste storage component 29; after water enters through the inlet, debris is trapped within the component 29 by the wastewater filters 30, while clean water is discharged from the pumps, thus achieving daily pool cleaning. Simultaneously, the thrust during drainage enhances the robot's friction on the bottom and its suction force when working against the pool wall.

[0160] In a further preferred embodiment of the present invention, such as Figure 16 As shown, the sensing mechanism 5 includes a sensing camera 27, at least one set of ultrasonic ranging sensors 28 and at least one set of supplementary lights 26. The ultrasonic ranging sensors 28 are detachably arranged on the outside of the robot body 1, and the supplementary lights 26 are arranged on both sides of the sensing camera 27. The sensing camera 27 is used to collect visual images of the pool wall and the pool bottom in real time.

[0161] In this embodiment, the number of ultrasonic ranging sensors 28 and supplementary lights 26 can be four sets. The four ultrasonic ranging sensors 28 are respectively arranged at the left front, right front, left, and right positions of the robot body 1 to realize wall distance detection. The four supplementary lights 26 are distributed on both sides of the sensing camera 27 to provide uniform and shadowless auxiliary lighting for the camera under low light conditions to ensure image acquisition quality. The sensing camera 27 is used to acquire visual images of the pool wall and bottom to identify stain types and terrain features. The sensing camera 27 should be placed in a waterproof housing, which is usually made of corrosion-resistant metal (such as stainless steel) or engineering plastic, and uses O-rings to achieve critical water tightness. Its protection level should not be lower than IP68 to ensure long-term stable operation at the predetermined working water depth.

[0162] In a further preferred embodiment of the present invention, such as Figure 17As shown, the control mechanism 7 includes a robot main control board, a storage tube 10, and a lighting switch 23. The robot main control board is electrically connected to the power supply unit 600. The lighting switch 23 is detachably installed in the frame stand 24. The frame stand 24 is detachably installed in the robot main frame 2. The storage tube 10 is installed in the frame stand 24, and a sealing flange 9 is fixedly installed at the end of the storage tube 10. At least one set of sealing threaded nuts 8 is provided in the sealing flange 9.

[0163] A wireless charging component 6 is located on the rear side of the robot's main frame 2. The wireless charging component 6 is electrically connected to the power supply unit 600 and contains a wireless charging induction coil. This wireless charging component 6 serves as the robot's power receiving unit, forming a magnetically coupled resonant wireless energy transmission system together with the power supply coil on the charging base station. The high-frequency alternating magnetic field generated by the base station penetrates the non-metallic isolation area of ​​the middle outer shell component 31, inducing an alternating current in the wireless charging component 6. This current then passes through internal rectification, filtering, and voltage conversion circuits, ultimately charging the lithium battery pack.

[0164] In this embodiment, the control mechanism 7 integrates a main control board with a microcontroller (MCU) or system-on-a-chip (SoC) as its core. This board houses a storage unit, power management circuitry, motor drive circuitry, and various communication interfaces (such as CAN, UART, PWM, etc.), collectively forming the hardware foundation for processing sensor information, executing control algorithms, and outputting commands. Furthermore, it can be expanded to include depth sensors, pH sensors, turbidity sensors, etc.

[0165] In summary, this invention provides a machine vision-based pool cleaning robot, control system, and method. In this embodiment, a machine vision-based pool cleaning robot control system is provided. This system, through the deep integration of machine vision technology and dynamic path planning algorithms, changes the traditional fixed-path or random cleaning mode of cleaning robots. The vision-based pool leaf recognition system can distinguish the type and distribution density of stains in real time. Combined with the dynamic path planning algorithm, the robot can autonomously identify highly polluted areas and prioritize their cleaning, achieving precise and personalized management of the cleaning process. This improves the robot's intelligence level and cleaning efficiency. Furthermore, the system can address the diverse pool environments (such as different materials for the pool bottom like tiles and mosaics) and slippage issues caused by smooth pool bottoms. It estimates the robot's underwater relative pose based on multi-directional ultrasonic distance parameters and yaw angles, and uses a slippage positioning method combining encoders and inertial sensors to correct the robot's position calculation. This effectively suppresses the accumulation of errors in the inertial navigation system and alleviates positioning distortion caused by wheel slippage; thus improving the robot's adaptability to complex environments and positioning accuracy.

[0166] It should be noted that, for the sake of simplicity, the foregoing embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to the present invention. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0167] It should be understood that the disclosed apparatus can be implemented in other ways, given the several embodiments provided in this application. For example, the apparatus embodiments described above are merely illustrative; the division of units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or communication connections shown or discussed may be through some interfaces; the indirect coupling or communication connections between devices or units may be telecommunications or other forms.

[0168] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit the scope of protection of the invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on these embodiments, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art can still combine, add, delete, or otherwise adjust the features of the various embodiments of the present invention according to the circumstances without conflict or creative effort, thereby obtaining different technical solutions that do not fundamentally depart from the concept of the present invention. These technical solutions also fall within the scope of protection of the present invention.

Claims

1. A machine vision based pool cleaning robot control system, characterized by: The machine vision-based pool cleaning robot control system includes: The sensing unit is used to collect multimodal data on the robot's internal state and external environment in real time; The main control unit is used to acquire multimodal data of the robot's internal state and external environment, preprocess the multimodal data, and output the preprocessed multimodal data. The decision-making unit generates motion planning and operation instructions based on the preprocessed multimodal data, and pushes the motion planning and operation instructions to the motion unit and the operation unit; The motion unit is used to receive instructions from the decision unit and parse the instructions into driving signals to drive the robot to perform movement, turning and climbing actions. The work unit is used to receive instructions from the decision unit, parse the instructions into work signals, and execute the pool cleaning operation; The power supply unit is electrically connected to the main control unit, decision-making unit, motion unit, and operation unit, and is used to supply power to the main control unit, decision-making unit, motion unit, and operation unit. The power supply unit includes a lithium battery pack, a charging management module, and a multi-channel voltage conversion circuit. The human-machine interaction unit is used to receive user commands and send them to the main control unit. At the same time, it receives and displays the operation status data transmitted back by the robot in real time. The human-machine interaction unit interacts with the main control unit through a wireless communication module. The main control unit includes: The data preprocessing unit is used to acquire multimodal data of the robot's internal state and external environment, preprocess the multimodal data, and output the preprocessed multimodal data. The relative pose estimation unit is used to acquire preprocessed multimodal data, identify multi-directional ultrasonic distance parameters and yaw angles in the multimodal data, estimate the robot's underwater relative pose based on the multi-directional ultrasonic distance parameters and yaw angles, and obtain the pool size parameters and the robot's relative position. The anti-slip positioning unit uses a slip positioning method based on the fusion of encoder and inertial sensor to calculate and correct the robot position, and outputs the corrected robot position. The underwater relative pose estimation method for the robot based on multi-directional ultrasonic distance parameters and yaw angle includes: Once the pool cleaning robot is powered on and enters the water at any location in the pool, it is driven forward by the motion unit to move close to any pool wall. After the pool cleaning robot first encounters the wall, it obtains the distance parameters between the robot and the pool wall or obstacle based on the ultrasonic ranging sensor. According to the distance parameters fed back by the two horizontally distributed ultrasonic ranging sensors, it adjusts the yaw angle of the pool cleaning robot so that the pool cleaning robot is perpendicular to the pool wall. Then the pool cleaning robot rotates 180° in any horizontal direction. Record the ultrasonic ranging sensor values ​​D1, D2, and D3 of the water tank cleaning robot in the forward, left, and right directions at this time. D1 is the average of the two forward ultrasonic ranging sensor values. Calculate the length and width of the water tank based on the ultrasonic ranging sensor values ​​D1, D2, and D3 of the water tank cleaning robot. The length of the water tank is D2 + D3, and the width is D1. A Cartesian coordinate system is established with one corner of the pool as a reference. The relative position of the pool cleaning robot is calculated based on the length and width of the pool, the yaw angle of the pool cleaning robot, and the values ​​D1, D2, and D3 of the forward, left, and right ultrasonic ranging sensors.

2. The machine vision-based pool cleaning robot control system as described in claim 1, characterized in that: The sensing unit includes: The visual perception module is used to acquire image information of the pool bottom and pool walls; The distance sensing module is used to measure the distance between the robot and the pool wall or obstacles in real time; The attitude perception module integrates a nine-axis inertial measurement unit (IMU) to detect the robot's own attitude in real time and provide feedback for motion control. The attitude perception module includes a gyroscope sensor, a geomagnetic sensor, and an accelerometer.

3. The machine vision-based pool cleaning robot control system as described in claim 2, characterized in that: The method for correcting robot position calculation based on the slippage positioning method using encoder and inertial sensor fusion includes: The raw data of the three-axis acceleration and three-axis angular velocity of the water tank cleaning robot are collected by the inertial measurement unit (IMU), and the raw data are filtered. The absolute heading angle information of the pool cleaning robot is obtained by the geomagnetic sensor. The rotational angular velocity of the drive wheel of the pool cleaning robot is measured by the encoder. The theoretical linear velocity or displacement is calculated based on the radius of the drive wheel. When the control command of the drive motor of the pool cleaning robot is detected to be zero and the position time exceeds the preset threshold, it is determined that the pool cleaning robot is in a stationary state and the zero speed correction process is triggered. A kinematic model of the robot is established, and a slip detection and adaptive fusion strategy is used to determine whether the pool cleaning robot is slipping. The slip detection and adaptive fusion strategy is as follows: the theoretical linear velocity calculated by the encoder is compared with the actual linear velocity calculated by the INS in real time. It is determined whether the absolute value of the theoretical linear velocity and the actual linear velocity exceeds a preset dynamic threshold. If the absolute value of the theoretical linear velocity and the actual linear velocity exceeds the preset dynamic threshold, the pool cleaning robot is determined to be slipping. If the absolute value of the theoretical linear velocity and the actual linear velocity does not exceed the preset dynamic threshold, the pool cleaning robot is determined to be in a normal state. When the pool cleaning robot is in a normal state, the data obtained by the encoder is used as the main observation source, and the actual linear velocity calculated by the INS is corrected based on the theoretical linear velocity calculated by the encoder. When the pool cleaning robot is determined to be slipping, the weight of the encoder data is reduced or completely removed, and the system is fused with the forward displacement observation of the INS after zero velocity correction and the lateral velocity constraint generated by the nonholonomic constraint, so that the system switches to the dead reckoning mode based on INS. The position solution is realized by an adaptive weight distribution mechanism: let the position solution of the INS be P i , the position solution of the encoder be P e , and the final position output be: ； ； Where α and β represent the first weighting coefficient and the second weighting coefficient; when the velocity difference When the dynamic threshold is exceeded, the second weighting coefficient is automatically adjusted so that the second weighting coefficient β approaches 0. At this point, the final position output is approximately... This ensures that positioning continuity is not affected by encoder distortion during slippage.

4. A machine vision-based pool cleaning robot control system as described in any one of claims 1-3, characterized in that, include: The robot body includes the robot's main frame; The walking mechanism, installed inside the robot's main frame, is used to assist the robot in moving and operating in the pool. The working mechanism, located within the main frame of the robot, is used to perform pool cleaning operations; The sensing mechanism is used to collect real-time environmental information of the pool and upload the information to the control mechanism. The control mechanism is used to acquire water tank environmental information uploaded by the sensing mechanism in real time, and trigger control commands for the operating mechanism based on the water tank environmental information.

5. The machine vision-based pool cleaning robot as described in claim 4, characterized in that: The robot's main frame includes a middle shell component, an upper shell component, a rear shell component, a lower shell component, a left shell component, and a right shell component. The upper shell component and the rear shell component are detachably connected to each other. The middle shell component is detachably connected to the upper shell component. The left shell component and the right shell component are symmetrically arranged on both sides of the middle shell component, and the left shell component and the right shell component are detachably connected to the middle shell component. The lower shell component is detachably connected to the left shell component and the right shell component. The walking mechanism includes a track drive motor, a track drive wheel, a track driven wheel, a transmission track, and a cross roller bearing. The track drive motor is fixedly installed inside the robot's main frame. The track drive wheel and the track driven wheel are rotatably installed on the robot's main frame and are rotatably connected to each other through the transmission track. The track driven wheel is rotatably connected to the robot's main frame through the cross roller bearing.

6. The machine vision-based pool cleaning robot as described in claim 5, characterized in that: The working mechanism includes a cleaning roller brush, a roller brush drive motor, a synchronous pulley drive wheel, a synchronous pulley driven wheel, a roller brush timing belt, a sludge storage component, and a water pump filtration mechanism. The cleaning roller brush is located on the front or lower side of the robot's main frame. The roller brush drive motor is fixedly installed inside the robot's main frame. The output shaft of the roller brush drive motor is fixedly connected to the synchronous pulley drive wheel. The synchronous pulley drive wheel and the synchronous pulley drive wheel are driven by the roller brush timing belt. The waste storage component is detachably installed inside the robot's main frame, and a wastewater inlet is provided at the bottom of the waste storage component. A wastewater filter screen is provided on the side wall of the waste storage component. The water pump filtration mechanism includes two sets of symmetrically arranged brushless power water pumps. The brushless power water pumps are used to provide power for the water flow during cleaning, and the brushless power water pumps are connected to the waste storage component.

7. The machine vision-based pool cleaning robot as described in claim 6, characterized in that: The sensing mechanism includes a sensing camera, at least one set of ultrasonic ranging sensors, and at least one set of supplementary lights. The ultrasonic ranging sensors are detachably arranged on the outside of the robot body, and the supplementary lights are set on both sides of the sensing camera. The sensing camera is used to collect visual images of the pool wall and the pool bottom in real time. The control mechanism includes a robot main control board, a storage tube, and a lighting switch. The robot main control board is electrically connected to the power supply unit. The lighting switch is detachably installed in the frame stand. The frame stand is detachably installed in the robot's main frame. The storage tube is installed in the frame stand, and a sealing flange is fixedly installed at the end of the storage tube. At least one set of sealing threaded nuts is installed in the sealing flange. A wireless charging component is provided on the rear side of the robot's main frame, and the wireless charging component is electrically connected to the power supply unit.

8. A machine vision-based control method for a pool cleaning robot, implemented based on the machine vision-based control system for a pool cleaning robot as described in any one of claims 1-3, characterized in that: The machine vision-based control method for the pool cleaning robot specifically includes: The pool environment information is collected by a sensing mechanism. The pool environment information includes image information and distance information. The pool environment information is enhanced and processed by adaptive histogram equalization and noise reduction to obtain the enhanced pool environment information. An object detection model integrating traditional vision and deep learning is used to analyze the enhanced pool environment information, calculate the target pixel ratio and distribution uniformity, and quantify the distribution of fallen leaves and sediment; the output includes the target object bounding box coordinates, center point pixel position and confidence score. By mapping the target pixel coordinates to the pool world coordinate system using camera calibration parameters and the current underwater relative pose of the pool cleaning robot; combined with the robot's odometry, the real-time relative distance and yaw angle between the pool cleaning robot and the target are obtained. Based on the feedback of the angle between the target position and the positive direction of the pool cleaning robot, a proportional-integral-derivative controller is designed to adjust the robot's steering. During the steering process, the target position is continuously updated until the robot is facing the target. The robot is controlled to move forward to the target location to perform cleaning, and the cleaning effect is verified in real time. Historical data of the robot's posture during the cleaning process is recorded. After the cleaning is completed, the robot is traced back to the starting point based on the historical status. During the process, the robot continuously identifies the target to be cleaned at the bottom of the pool until no target is detected.

Images