Sweeping robot cleaning control system with self-adaptive mode adjustment

Abstract

The invention discloses a sweeping robot cleaning control system with a self-adaptive mode adjustment function, and relates to the technical field of intelligent cleaning robot control systems, and the system comprises a multi-mode sensing module, a central processing module and a cleaning execution module. The multi-modal sensing module collects multi-dimensional data such as a hyperspectral image, acoustic response and gas concentration of ground stains, and the central processing module extracts features through the data fusion unit and deeply fuses the features to generate analysis results of stain types, adhesion strength and chemical components. The decision control unit dynamically generates a cleaning instruction in combination with a preset strategy and the real-time state of the robot, the cleaning execution module executes a targeted cleaning action, and meanwhile, the system constructs a dynamic stain atlas to achieve historical data reuse and prospective adjustment, so that the cleaning efficiency and the resource utilization efficiency are remarkably improved, and the cleaning effect is improved. The method is suitable for intelligent cleaning scenes of home and office environments.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of control systems of intelligent cleaning robots, in particular to a cleaning control system of a sweeping robot with adaptive mode adjustment. BACKGROUND

[0002] With the improvement of people's living standards and the rapid development of smart home technology, sweeping robots have become an indispensable cleaning tool for many families. Its development process has gone through random collision cleaning, planning cleaning based on gyroscopes and acceleration sensors, and current widely used navigation and obstacle avoidance based on laser radar (LIDAR) or visual simultaneous localization and mapping (SLAM) technology. The degree of intelligence is constantly improving. Market demand has evolved from basic floor cleaning to deeper cleaning effect, higher cleaning efficiency, lower energy consumption, and more personalized cleaning experience. However, the current sweeping robot still has a significant bottleneck in the adaptive adjustment of its cleaning mode when facing complex and variable home environments and stain types.

[0003] Existing sweeping robot cleaning control systems have tried to introduce various technologies to improve adaptability. For example, the invention patent application of Suzhou University; Household cleaning robot control system based on adaptive strategy optimization (application number: 201810199176.2) discloses a method using reinforcement learning algorithm. The system obtains environmental information through the sensing system, uses the strategy network to control the action of the robot, and uses the value network to evaluate the action, thereby realizing adaptive control. This method is an improvement compared to traditional fixed rule control strategy, but its perception dimension is still limited, and its environmental information is mostly concentrated in visual (such as camera) and non-visual (such as ultrasonic, infrared) physical space information, such as obstacles, cliffs, carpets (identified by ultrasonic waves), etc. Another common solution, such as the solution applied in the product of Shitetechnology, automatically switches to the only sweeping mode after identifying the carpet through the ultrasonic sensor, or sets different suction power and water volume according to the room type. These technologies are still essentially a response to physical space and preset rules, and cannot identify and understand the chemical composition and physical properties (such as viscosity, water content) of stains themselves, resulting in difficulty in generating truly optimal and targeted cleaning parameter combinations when facing stains of different properties (such as oil stains, sugar stains, dust), often causing resource waste due to "overcleaning" or leaving stains due to "insufficient cleaning".

[0004] Therefore, the problem to be solved is how to break through the limitation of the prior art that mainly relies on geometric space information and simple physical signals (such as ultrasonic echoes) for mode switching, so that the sweeping robot can accurately perceive and analyze the chemical composition and physical properties of the ground stains, and accurately and adaptively adjust the cleaning mode based thereon. The prior art lacks effective detection means for stain composition, and further fails to deeply couple the composition information with the cleaning strategy, which is the fundamental reason why it cannot realize truly intelligent and refined cleaning.

[0005] In summary, the cleaning control system of the existing sweeping robot has obvious deficiencies in the perception dimension and decision intelligence. Most of the schemes focus on the adaptation to the spatial layout and the macro terrain (such as the floor and the carpet), but lack the ability to deeply perceive and understand the micro characteristics of the cleaning object, i.e. the stain itself. This greatly limits the further improvement of cleaning efficiency and the optimization of resource allocation. Developing a system that can accurately analyze the composition of the stain in real time and dynamically generate the optimal cleaning strategy has great significance for breaking through the technical bottleneck of the industry and meeting the needs of the high-end market. SUMMARY

[0006] The purpose of the present application is to make up for the deficiencies of the prior art, and provide a sweeping robot cleaning control system with adaptive mode adjustment. Through the cooperative architecture of the multi-modal sensing module, the central processing module and the cleaning execution module, the adaptive mode adjustment of the sweeping robot cleaning control is realized. The multi-modal sensing module synchronously collects multi-dimensional data such as hyperspectral images, acoustic responses and gas concentrations of ground stains, providing comprehensive and original information for stain characteristic analysis. The central processing module extracts features and deeply fuses through the data fusion unit, generates analysis results of stain type, adhesion strength and chemical composition, and the decision control unit dynamically optimizes the cleaning instruction in combination with the preset strategy and the real-time state of the robot. The cleaning execution module executes the matching action, and at the same time constructs a dynamic stain map to realize data reuse, thereby making up for the deficiencies of the prior art and improving the cleaning accuracy and efficiency.

[0007] To solve the above technical problems, the present application provides the following technical scheme: a sweeping robot cleaning control system with adaptive mode adjustment, comprising:

[0008] a multi-modal sensing module, a central processing module and a cleaning execution module;

[0009] The multi-modal sensing module is used to collect multi-dimensional original data of the environment and ground stains where the robot is located, and includes a hyperspectral imaging unit, an acoustic resonance analysis unit and a chemical sensing unit;

[0010] The hyperspectral imaging unit is configured to emit a wide-spectrum light beam to a target area on the ground and receive a reflected light signal to obtain hyperspectral image data of the target area.

[0011] The acoustic resonance analysis unit is configured to emit an acoustic signal of a specific frequency to the target area on the ground and receive an acoustic echo signal reflected by the target area on the ground to obtain acoustic response data characterizing physical properties of the stain.

[0012] The chemical sensing unit is configured to adsorb and detect gas molecules volatilized from the target area on the ground to obtain gas concentration data characterizing chemical components of the stain.

[0013] The central processing module is electrically connected with the multi-modal sensing module and is configured to receive and process the multi-dimensional raw data, including a data fusion unit and a decision control unit.

[0014] The data fusion unit is configured to pre-process, extract features, and perform fusion calculation on the received hyperspectral image data, acoustic response data, and gas concentration data to generate a comprehensive stain component analysis result including at least a stain type identifier, an adhesion strength quantitative value, and a chemical component identifier.

[0015] The decision control unit is connected with the data fusion unit and is configured to query a pre-stored cleaning strategy mapping table based on the stain component analysis result to generate a corresponding cleaning mode control instruction set.

[0016] The cleaning execution module is electrically connected with the central processing module and is configured to receive the cleaning mode control instruction set and execute cleaning actions defined by the control parameters.

[0017] Further, the data fusion unit includes a hyperspectral feature extraction subunit, an acoustic feature extraction subunit, and a chemical feature extraction subunit.

[0018] The hyperspectral feature extraction subunit is configured to perform dimension reduction and denoising processing on the hyperspectral image data and extract a spectral feature vector matching a known stain sample library.

[0019] The acoustic feature extraction subunit is configured to perform frequency domain transformation on the acoustic response data to extract energy attenuation features and resonance frequency shift features in a specific frequency band to form an acoustic feature vector.

[0020] The chemical feature extraction subunit is configured to perform normalization processing on the gas concentration data and identify one or more types of volatile organic compounds and their concentration ratios based on a sensor response curve to form a chemical feature vector.

[0021] The data fusion unit further comprises a feature fusion subunit based on a deep neural network, which receives the spectral feature vector, the acoustic feature vector and the chemical feature vector, and performs weighted fusion and joint classification through a trained multi-branch neural network model to finally output the stain component analysis result.

[0022] Further, the multi-branch neural network model based on a deep neural network in the feature fusion subunit realizes the final fusion and classification decision through an optimization objective function, which is defined as:

[0023]

[0024]

[0025] wherein, represents the number of sample data in a training batch, represents the total number of stain types, is a binary indicator function, indicating whether the real class of the sample is 1 when the real class of the sample is , otherwise 0, represents the probability that the multi-branch neural network model predicts that the sample belongs to the class , represents the weight matrix of the hyperspectral feature extraction subunit branch network, represents the weight matrix of the acoustic feature extraction subunit branch network, represents the weight matrix of the chemical feature extraction subunit branch network, represents an L2 norm regularization function applied to the weight matrix, , , and are regularization hyperparameters corresponding to the weight matrix of each branch network, used to control the model complexity and prevent overfitting.

[0026] Further, the decision control unit comprises:

[0027] The cleaning mode control instruction set at least includes regulation and control parameters of the suction motor speed, the water tank water pump flow, the roller brush speed and the vibration frequency;

[0028] The strategy mapping subunit and the dynamic optimization subunit;

[0029] The strategy mapping subunit stores the cleaning strategy mapping table, which defines the corresponding relationship between different types of stain component analysis results and basic cleaning mode parameters;

[0030] The dynamic optimization subunit is configured to receive real-time state data of the robot, the real-time state data including a remaining battery capacity, a remaining clean water tank capacity, and a current sewage tank capacity.

[0031] The dynamic optimization subunit is further configured to perform online adjustment on the basic cleaning mode parameters output by the strategy mapping subunit according to the real-time state data, so as to achieve optimal allocation of cleaning performance under system resource constraints, and output a final cleaning mode control instruction set.

[0032] Furthermore, when the dynamic optimization subunit performs online adjustment on the basic cleaning mode parameters according to the real-time state data, a multi-objective optimization problem is solved as follows:

[0033] wherein, represents a cleaning mode control instruction sequence for a future time steps from a current time point, represents a length of an optimization time domain, represents a total energy consumption cost function related to a cleaning action in the optimization time domain, represents a total water consumption cost function in the optimization time domain, represents a system instantaneous power determined by a control instruction at a first time step, represents a water pump instantaneous flow determined by a control instruction at a second time step, and are weight coefficients of energy consumption and water consumption, respectively, for adjusting relative importance of two optimization objectives, represents a total remaining battery capacity, represents a battery capacity safety threshold allowed by the system, represents a total remaining clean water tank capacity, represents a clean water tank safety threshold allowed by the system, and represent upper and lower limit constraints of the cleaning mode control instruction, represents a time interval of one time step.

[0034] Furthermore, the sound wave resonance analysis unit includes a sound wave transmitter array and a sound wave receiver array.

[0035] The sound wave transmitter array is composed of a plurality of piezoelectric ceramic transducers capable of independently emitting sound waves of different frequencies.

[0036] The sound wave receiver array is composed of a plurality of microphone sensors for synchronously receiving the sound wave echo signals reflected from different angles;

[0037] The sound wave resonance analysis unit calculates the acoustic impedance characteristics and viscoelastic modulus of the stain as an important part of the acoustic response data by controlling the sound wave transmitter array to emit frequency scanning signals in a certain time sequence and collecting echoes via the sound wave receiver array.

[0038] Further, the chemical sensing unit adopts a solid phase microextraction-gas chromatography mass spectrometry technology module;

[0039] It includes a micro sampling pump, a solid phase microextraction fiber head, a micro heating desorption cavity, and a metal oxide semiconductor gas sensor array;

[0040] The micro sampling pump is used to extract and blow the gas near the ground through the solid phase microextraction fiber head;

[0041] The solid phase microextraction fiber head is used to adsorb and enrich volatile organic molecules in the gas;

[0042] The micro heating desorption cavity is used to heat the solid phase microextraction fiber head after adsorption saturation to desorb the volatile organic molecules;

[0043] The metal oxide semiconductor gas sensor array is exposed to the desorbed gas to generate electrical signal changes corresponding to specific chemical groups, thereby generating the gas concentration data.

[0044] Further, the central processing module further includes a dynamic stain map modeling unit;

[0045] The dynamic stain map modeling unit is in communication connection with the data fusion unit and the navigation and positioning module of the robot;

[0046] The dynamic stain map modeling unit is configured to receive the stain component analysis result from the data fusion unit and the stain geographic location information from the navigation and positioning module;

[0047] The dynamic stain map modeling unit binds the stain component analysis result and the stain geographic location information and records the time stamp, thereby constructing and real-time updating a dynamic stain map on the environmental map of the robot;

[0048] The dynamic stain map not only marks the spatial distribution of the stain, but also stores the historical component analysis result sequence of each stain point in a hierarchical data structure.

[0049] Further, the decision control unit is also connected with the dynamic stain map modeling unit;

[0050] The decision control unit is configured to, when generating the cleaning mode control instruction set, not only according to the stain component analysis result generated by the current data fusion unit in real time, but also call the historical component analysis result sequence of the corresponding geographical position in the dynamic stain map;

[0051] The decision control unit predicts the evolution direction of the stain state by analyzing the change trend of the historical component analysis result sequence, and makes forward-looking adjustment to the cleaning mode control instruction set.

[0052] Further, the cleaning execution module includes a suction force stepless speed regulation motor, a water pump flow control motor and a rolling brush torque adjustable motor;

[0053] The suction force stepless speed regulation motor is configured to receive the instruction about the speed of the suction motor in the cleaning mode control instruction set, and drive the fan to realize continuous adjustment of the suction force;

[0054] The water pump flow control motor is configured to receive the instruction about the water tank water pump flow in the cleaning mode control instruction set, and drive the micro gear pump to realize accurate control of the water output;

[0055] The rolling brush torque adjustable motor is configured to receive the instruction about the rolling brush speed and vibration frequency in the cleaning mode control instruction set, and drive the rolling brush assembly to operate at a specified speed and torque mode, wherein the vibration frequency is realized by changing the pulse frequency of the motor input current.

[0056] Compared with the prior art, the self-adaptive mode adjusting sweeping robot cleaning control system has the following beneficial effects:

[0057] Firstly, the present application synchronously collects multi-dimensional data such as hyperspectral images, acoustic responses and gas concentrations through a multi-modal sensing module, and performs feature extraction and deep fusion through the data fusion unit of the central processing module, which can accurately analyze the stain type, adhesion strength and chemical composition.

[0058] Secondly, the application realizes historical data reuse and forward regulation by constructing a dynamic stain map, greatly improves cleaning efficiency and resource utilization efficiency, and records information such as stain position, composition and cleaning effect in real time and updates the dynamic stain map, thereby providing data support for subsequent cleaning tasks; the decision control unit optimizes cleaning parameters in combination with historical trends in the map and real-time power and water quantity states of the robot, thereby reducing unnecessary energy consumption and water consumption; meanwhile, the forward regulation mechanism enables the system to actively adapt to environmental changes, reduces the probability of occurrence of a cleaning blind area, improves the utilization efficiency of the battery and water resources under the premise of ensuring cleaning quality, and enhances the adaptability and economy of the system in long-term use.

[0059] Additional advantages, objects, and features of the application will be set forth in part by the description that follows, and in part will become apparent to those skilled in the art upon examination of same, or can be learned by practice of the application. The objects and advantages of the application can be realized and attained by means of the instrumentalities and combinations particularly pointed out in the appended claims. BRIEF DESCRIPTION OF DRAWINGS

[0060] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiment or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0061] Figure 1 The operation flowchart of the present application;

[0062] Figure 2 The multi-modal sensing and feature fusion structure diagram of the present application;

[0063] Figure 3 The dynamic decision mechanism diagram under resource constraints of the present application. DETAILED DESCRIPTION

[0064] In order to further illustrate the technical means and effects adopted by the present application to achieve the predetermined application purposes, the specific embodiments, structures, features and effects according to the present application are described in detail below in combination with the drawings and preferred embodiments.

[0065] Embodiment one

[0066] As Figure 2 and Figure 3As shown, the embodiment discloses a self-adaptive mode adjustment of a sweeping robot cleaning control system, aiming to solve the technical problems of insufficient perception of stain micro features and low cleaning mode adjustment precision of existing sweeping robots. The system collects multi-dimensional data such as hyperspectral images, acoustic responses and gas concentrations of ground stains through a multi-modal sensing module, extracts and deeply fuses features through the data fusion unit of the central processing module, and generates analysis results containing stain type, adhesion strength and chemical composition; the decision control unit dynamically optimizes and generates cleaning mode control instructions based on the results combined with the preset strategy mapping table and the real-time state of the robot, and finally executes targeted cleaning actions by the cleaning execution module. At the same time, the system realizes historical data reuse and forward-looking adjustment by building a dynamic stain map, significantly improving cleaning efficiency and resource utilization efficiency, and is suitable for intelligent cleaning scenes in various home and office environments.

[0067] The implementation process of the overall system architecture and workflow is as follows:

[0068] The self-adaptive mode adjustment of the sweeping robot cleaning control system adopts modular design, mainly composed of a multi-modal sensing module, a central processing module and a cleaning execution module, and data transmission and instruction interaction are realized between modules through electrical connection. The overall workflow follows the closed-loop control logic of perception-analysis-decision-execution.

[0069] In actual application, when the sweeping robot starts a cleaning task, the multi-modal sensing module first scans the ground area in the travel path in real time, synchronously collecting hyperspectral image data, acoustic response data and gas concentration data. These data, as raw information representing stain characteristics, will be continuously transmitted to the central processing module. After receiving the data, the central processing module first preprocesses and extracts features through the data fusion unit, and then uses a deep neural network model to complete the fusion and classification of multi-dimensional features, generating analysis results containing stain type, adhesion strength quantitative value and chemical composition identification. Based on the analysis results, the decision control unit preliminarily determines the basic cleaning parameters combined with the preset cleaning strategy mapping table, and then dynamically optimizes the parameters according to the real-time state data of the robot such as battery power and water quantity, forming the final cleaning mode control instruction set. After receiving the instructions, the cleaning execution module adjusts the operating parameters of the suction motor, water pump and brush motor to execute cleaning actions matching the stain characteristics. At the same time, the system records information such as stain location, composition and cleaning effect in real time, updates the dynamic stain map, and provides data support for subsequent cleaning tasks.

[0070] The specific implementation of the multi-modal sensing module is as follows:

[0071] The multi-modal sensing module is the core of the system to achieve accurate perception. Through the collaborative work of the hyperspectral imaging unit, the sound wave resonance analysis unit and the chemical sensing unit, it captures the characteristic information of the stain from the three dimensions of optics, physics and chemistry, providing a comprehensive data basis for subsequent analysis.

[0072] The hyperspectral imaging unit is mainly used to obtain the spectral characteristics of the target area, which is composed of a wide-spectrum light source, a light splitting element and an imaging sensor. When working, the wide-spectrum light source emits a wide-spectrum light beam covering the visible light to near-infrared band to the target area on the ground. After reflection, the reflected light is decomposed by the light splitting element according to the wavelength, and then the imaging sensor receives the light signals of different wavelengths to generate hyperspectral image data containing spatial information and spectral information. By analyzing the reflectivity difference of different stains to specific wavelengths of light, the physical state (such as liquid, solid) and surface characteristics (such as glossiness, roughness) of the stain can be preliminarily distinguished. For example, the reflectivity of oil stains to near-infrared band is usually lower than that of water stains, while dust stains will form obvious reflection peaks at certain bands. These characteristics provide optical basis for subsequent stain type identification.

[0073] The sound wave resonance analysis unit is used to obtain acoustic response data representing the physical characteristics of the stain, which is composed of a sound wave transmitter array, a sound wave receiver array and a signal processing circuit. The sound wave transmitter array includes multiple piezoelectric ceramic transducers, each of which can independently emit a specific frequency sound wave in the range of 20 kHz to 200 kHz; the sound wave receiver array is composed of multiple microphone sensors, which can synchronously receive reflected sound wave echo signals from different angles. When working, the system controls the transmitter array to emit frequency scanning signals according to the preset timing. After the sound wave acts on the ground stain, part of the energy is absorbed or scattered by the stain, and the remaining energy forms a reflected echo. After the signal processing circuit amplifies and filters the echo signal, the energy attenuation characteristics and resonance frequency shift characteristics of the sound wave during propagation are extracted. By analyzing these characteristics, the acoustic impedance characteristics and viscoelastic modulus of the stain can be calculated: the acoustic impedance reflects the hindering ability of the stain to sound wave propagation, and the viscoelastic modulus represents the adhesion strength and fluidity of the stain. For example, viscous sauce stains will cause significant attenuation of sound wave energy and large resonance frequency shift, while dry dust stains have less effect on sound waves.

[0074] The chemical sensing unit is used for detecting volatile organic compounds volatilized from stains to realize recognition of chemical components. The chemical sensing unit is composed of a micro sampling pump, a solid-phase micro extraction fiber head, a micro heating desorption cavity and a metal oxide semiconductor gas sensor array according to the principle of solid-phase micro extraction-gas chromatography mass spectrometry. During operation, the micro sampling pump continuously extracts and blows the gas near the ground over the solid-phase micro extraction fiber head. The coating on the surface of the fiber head can selectively adsorb volatile organic molecules in the gas. After enrichment for a preset time, the fiber head is transferred to the micro heating desorption cavity. The desorption cavity increases the temperature of the fiber head to a preset value through resistance heating, so that the adsorbed organic molecules are desorbed and released. The desorbed gas enters the metal oxide semiconductor gas sensor array. The metal oxide material on the surface of the sensor reacts with specific chemical groups, causing a change in electrical conductivity. By detecting the change in the electrical signal, the type (such as alkanes, esters, aldehydes, etc.) and concentration ratio of the organic matter can be identified. For example, fatty acid methyl esters volatilized from kitchen oil stains can significantly reduce the electrical conductivity of a specific sensor, while sugar decomposition in beverage stains can cause another sensor to change signals. These chemical characteristics provide a key basis for distinguishing between oil stains, juice stains and sauce stains.

[0075] The data fusion unit of the central processing module implements the following;

[0076] The data fusion unit is the core of multi-dimensional data analysis. Through the cooperative work of the hyperspectral feature extraction subunit, the acoustic feature extraction subunit, the chemical feature extraction subunit and the feature fusion subunit, the original sensing data is converted into stain component analysis results that can be used for decision making, solving the problem of incomplete information from a single data source.

[0077] The hyperspectral feature extraction subunit is responsible for preprocessing and feature extraction of hyperspectral image data. First, principal component analysis algorithm is used to reduce the dimension of hyperspectral image data, remove redundant spectral bands, retain characteristic bands containing main information, and reduce data processing complexity. At the same time, Gaussian filtering is used to remove random noise in the image and enhance the signal-to-noise ratio of the image. Subsequently, the subunit compares the processed image data with a pre-set known stain sample library, extracts the spectral feature vector matching the sample library, and the vector contains key information such as reflectivity and absorption peak position of the stain at different characteristic bands, providing a basis for preliminary classification of stain types.

[0078] The acoustic feature extraction subunit extracts features from the acoustic response data. First, the acoustic echo signal in the time domain is converted to the frequency domain by the fast Fourier transform to obtain the frequency distribution characteristics of the signal. Then, the energy attenuation feature and the resonance frequency shift feature are extracted in the preset feature frequency band (e.g., 50 kHz to 150 kHz). The energy attenuation feature is obtained by calculating the signal energy loss rate in a specific frequency range, and the resonance frequency shift feature is determined by comparing the difference between the transmission frequency and the main peak frequency in the echo signal. These features are integrated to form an acoustic feature vector, which quantitatively reflects the physical properties of the stain, such as viscoelasticity and density. For example, a stain with high adhesion strength has a larger energy attenuation feature value and a more obvious resonance frequency shift.

[0079] The chemical feature extraction subunit is used to extract a chemical feature vector from the gas concentration data. First, the electrical signals output by the sensor array are normalized to eliminate the effects of different sensor sensitivities and map the signal values uniformly to the interval of 0-1. Then, according to the pre-set sensor response curve (which is calibrated by a large number of experiments and records the corresponding relationship between different organic matter concentrations and sensor electrical signals), one or more types of volatile organic compounds contained in the gas are identified, and the concentration proportion of each type is calculated. For example, when the concentration proportion of ester substances is detected to be more than 60%, it can be preliminarily determined that the stain may be an oil stain; if a high concentration of sugar decomposition products is also detected, it can be further inferred that it is an oil-containing food residue stain. These information is integrated to form a chemical feature vector, which provides a basis for identifying the chemical composition of the stain.

[0080] The feature fusion subunit uses a multi-branch neural network model based on deep neural networks to realize the fusion and classification of features. The model includes three parallel feature processing branches (corresponding to the spectral, acoustic, and chemical feature vectors, respectively) and a fusion classification layer. First, the spectral feature vector, acoustic feature vector, and chemical feature vector are input into the corresponding branch network, and each branch network processes the features through convolutional layers and fully connected layers to extract higher-order features with better discrimination. Then, the fusion classification layer performs weighted fusion on the high-order features output by the three branches, and the weight values are dynamically adjusted through model training to highlight the features that contribute more to classification.

[0081] The training and optimization of the model are realized by optimizing the objective function, which is defined as:

[0082]

[0083]

[0084] wherein, N represents the number of sample data in a training batch, N represents the total number of stain types, is a binary indicator function indicating whether the true class of a sample is 1 if the true class of a sample is 1, otherwise 0, denotes the probability that the multi-branch neural network model predicts that a sample belongs to class , denotes the weight matrix of the hyperspectral feature extraction subunit branch network, denotes the weight matrix of the acoustic wave feature extraction subunit branch network, denotes the weight matrix of the chemical feature extraction subunit branch network, denotes the L2 norm regularization function applied to the weight matrix, , , are regularization hyperparameters corresponding to the weight matrices of the branch networks, respectively, used to control the model complexity and prevent overfitting. By minimizing the objective function, the excessive growth of the weight matrices can be suppressed while ensuring the classification accuracy of the model, thereby reducing the risk of overfitting. After training is completed, the model can output comprehensive analysis results including stain type identification, adhesion strength quantification value, and chemical composition identification.

[0085] The decision control unit of the central processing module implements the following;

[0086] The decision control unit is the core of the system for intelligent adjustment. Through the collaborative work of the strategy mapping subunit and the dynamic optimization subunit, it converts the stain component analysis results into cleaning mode control instructions that adapt to the state of the robot, ensuring the balance between cleaning efficiency and resource consumption.

[0087] The cleaning mode control instruction set output by the decision control unit covers key parameters such as suction motor speed, water tank water pump flow, roller brush speed, and vibration frequency. The combination of these parameters directly determines the intensity and mode of cleaning action. The strategy mapping subunit has a pre-installed cleaning strategy mapping table, which is constructed through a large amount of experimental data and defines the correspondence between different stain component analysis results and basic cleaning parameters. For example, for oil stains with high adhesion strength, the mapping table matches high suction (high motor speed), moderate water volume (moderate pump flow), high roller brush speed, and high vibration frequency basic parameters; for dust with low adhesion strength, the mapping table corresponds to low suction, low water volume, moderate roller brush speed, and low vibration frequency parameter combination. When receiving the stain component analysis results, the strategy mapping subunit quickly determines the basic cleaning mode parameters by looking up the table.

[0088] The dynamic optimization subunit is used to adjust the basic parameters online according to the real-time state of the robot, and the data received by the subunit include the remaining battery capacity, the remaining water in the clean water tank, and the current capacity of the sewage tank. The core of the subunit is to solve a multi-objective optimization problem to achieve optimal allocation of cleaning performance under system resource constraints. The optimization objective function is defined as:

[0089] wherein, represents the cleaning mode control instruction sequence for the future time steps from the current time, represents the length of the optimization time domain, represents the total energy consumption cost function related to the cleaning action in the optimization time domain, represents the total water consumption cost function in the optimization time domain, represents the system instantaneous power determined by the control instruction at the time step, represents the water pump instantaneous flow determined by the control instruction at the time step, and are the weight coefficients of energy consumption and water consumption, respectively, for adjusting the relative importance of the two optimization objectives, represents the total remaining battery capacity, represents the battery capacity safety threshold allowed by the system, represents the total remaining water in the clean water tank, represents the clean water safety threshold allowed by the system, and represent the upper and lower limit constraints of the cleaning mode control instruction, represents the time interval of one time step.

[0090] In actual optimization, the dynamic optimization subunit adjusts the basic cleaning parameters by solving the above optimization problem through a gradient descent algorithm according to the current power and water state. For example, when the battery capacity is low, the subunit will appropriately reduce the suction motor speed to reduce energy consumption while maintaining the effective cleaning intensity of the roller brush; when the clean water is insufficient, the water pump flow will be reduced to prioritize spot cleaning of stubborn stains. The adjusted parameters form the final cleaning mode control instruction set, which is transmitted to the cleaning execution module.

[0091] The specific implementation of the cleaning execution module is as follows:

[0092] The cleaning execution module is the execution terminal of the system instruction, which converts the instruction output by the decision control unit into specific cleaning actions through the precise control of the suction motor, the water pump flow control motor and the roll brush torque adjustable motor, so as to realize the cleaning effect matched with the characteristics of the stains.

[0093] The suction motor is composed of a direct-current brushless motor and a fan. After receiving the speed instruction in the cleaning mode control instruction set, the motor controller adjusts the input voltage of the motor through pulse width modulation technology to realize continuous adjustment of the speed, and then changes the suction force generated by the fan. When facing stains with high adhesion strength, the motor runs at high speed, and the fan generates strong suction force to ensure that the stains are effectively sucked into the dust collection box. When cleaning ordinary dust, the motor runs at low speed to reduce energy consumption and dust raising. During the operation of the motor, the built-in Hall sensor feedbacks the speed information in real time, and the controller ensures that the actual speed is consistent with the instruction speed through closed-loop control.

[0094] The water pump flow control motor is composed of a micro gear pump and a stepping motor, which is used to accurately control the water output of the clean water tank. After receiving the flow instruction, the stepping motor rotates at the preset step frequency through the drive circuit, driving the gear of the gear pump to mesh and rotate, and pumping clean water from the water tank to the ground. The flow of the gear pump is linearly related to the number of steps of the motor, and the flow can be accurately adjusted by controlling the number of steps of the stepping motor. For example, when cleaning oil stains, more water is needed to cooperate with the cleaning agent to dissolve the stains, the number of steps of the motor is increased, and the water output is increased; when cleaning dry stains, the number of steps is reduced to reduce the water output to avoid leaving too much water on the ground.

[0095] The roll brush torque adjustable motor is composed of a direct-current motor and a speed reduction gear set, which is used to drive the roll brush assembly to run at a specified speed and vibration frequency. After receiving the speed and vibration frequency instructions, the motor controller controls the basic speed of the motor by adjusting the input voltage on one hand, and controls the vibration of the roll brush by changing the pulse frequency of the input current on the other hand: the higher the pulse frequency, the higher the vibration frequency of the roll brush, and the stronger the effect of stripping stubborn stains. At the same time, the torque sensor built-in the motor monitors the load change in real time, and when the roll brush encounters hair entanglement or other increased resistance, the controller will automatically increase the output torque to avoid motor stall and ensure the continuity of the cleaning process. For example, when cleaning deep stains on the carpet, the motor runs at high speed and high vibration frequency to strip the stains from the fibers through mechanical friction and vibration of the roll brush; when cleaning hard floors, the speed and vibration frequency are reduced to reduce the wear on the floor.

[0096] The dynamic stain map modeling and application are implemented as follows:

[0097] The dynamic stain map modeling unit is the key to realize the reuse of historical data and forward-looking adjustment. It records and analyzes the spatial and temporal distribution and component variation of stains to provide long-term data support for decision control, thereby improving the self-adaptive ability of the system.

[0098] The dynamic stain map modeling unit is in communication connection with the data fusion unit and the navigation positioning module. The navigation positioning module obtains the real-time position information of the robot through laser radar or visual SLAM technology, thereby providing a spatial coordinate reference for stain positioning. In operation, the unit receives the stain component analysis results from the data fusion unit in real time, including stain type, adhesion strength, chemical composition and the like, and receives the stain geographical position coordinates (such as room area, distance from the starting point, etc.) output by the navigation positioning module, and adds a time stamp (accurate to seconds) to each piece of data. Subsequently, the unit binds the stain component analysis results and the geographical position information, and stores them in a hierarchical data structure: the first layer is a spatial distribution layer, which marks the position coordinates of each stain point in the form of a two-dimensional map; the second layer is a component attribute layer, which records the type, strength and chemical composition identifier of each stain point; and the third layer is a historical time sequence layer, which stores the stain component analysis results detected at the same position multiple times in chronological order, thereby forming sequence data.

[0099] When generating the cleaning mode control instruction set, the decision control unit calls the historical data of the corresponding geographical position in the dynamic stain map. By analyzing the change trend of the historical component analysis result sequence, the evolution direction of the stain state can be predicted: for example, if oil stains are detected multiple times in a corner of the kitchen and the adhesion strength gradually increases, it indicates that there may be persistent oil stain pollution in this area, and the decision control unit will increase the suction force and water volume parameters of this area in subsequent cleaning; if dust stains frequently occur in a certain area at a specific time every week, the system will adjust the cleaning plan and increase the cleaning frequency of this area during this time period. This forward-looking adjustment mechanism enables the system to actively adapt to environmental changes and reduce cleaning blind spots and resource waste.

[0100] In summary, the present embodiment realizes the precise perception of stain characteristics and the adaptive adjustment of the cleaning mode of the robot through the collaborative design of the multi-modal sensing module, the central processing module and the cleaning execution module. The multi-modal sensing module comprehensively captures the stain characteristics from the optical, physical and chemical dimensions, thereby solving the problem of insufficient information of a single sensing method; the data fusion unit realizes efficient fusion of multi-dimensional features through a deep neural network, thereby improving the accuracy of stain analysis; the decision control unit dynamically optimizes the pre-set strategy and real-time state, thereby balancing the cleaning efficiency and resource consumption; the precise control of the cleaning execution module ensures the effective landing of the instructions; and the construction and application of the dynamic stain map realize the reuse of historical data and forward-looking adjustment, thereby further improving the intelligent level of the system.

[0101] Compared with the prior art, the system can generate customized cleaning strategies for different types and intensities of stains, effectively reducing over-cleaning or under-cleaning, improving cleaning effect while reducing energy consumption and water consumption. The implementation of the system provides a feasible solution for the technical upgrade of intelligent cleaning robots, has strong practicality and popularization value, and is suitable for floor cleaning needs in various scenes such as home, office, and business.

[0102] Embodiment Two

[0103] As shown in the embodiment one, on the basis of the embodiment one, the embodiment two elaborates the specific steps of the cleaning control system of the sweeping robot in the adaptive mode adjustment when working, and the specific steps are as follows: Figure 1

[0104] 1. System startup and initialization

[0105] The sweeping robot is powered on, each sensor and actuator is self-checked and initialized, the central processing module loads the preset cleaning strategy mapping table and multi-branch neural network model, and the dynamic stain map modeling unit initializes the environment map.

[0106] 2. Multi-modal data acquisition

[0107] In the process of moving, the robot synchronously scans the current ground area through the multi-modal sensing module:

[0108] The hyperspectral imaging unit emits a wide spectrum light beam and receives reflected light to generate hyperspectral image data;

[0109] The acoustic resonance analysis unit emits a specific frequency sound wave and receives the echo to obtain acoustic response data;

[0110] The chemical sensing unit adsorbs and detects volatile gases to generate gas concentration data.

[0111] 3. Data transmission and preprocessing

[0112] The multi-modal sensing module uploads the collected hyperspectral image data, acoustic response data and gas concentration data to the data fusion unit of the central processing module, and each feature extraction subunit performs noise reduction, normalization, frequency domain transformation and other preprocessing operations.

[0113] 4. Feature extraction and fusion

[0114] The hyperspectral feature extraction subunit extracts a spectral feature vector;

[0115] The acoustic feature extraction subunit extracts an acoustic feature vector;

[0116] The chemical feature extraction subunit extracts a chemical feature vector;

[0117] ​The feature fusion subunit performs weighted fusion and joint classification on the three types of feature vectors through a multi-branch neural network model, and outputs the stain component analysis result, including stain type, adhesion strength and chemical composition.

[0118] 5. Cleaning strategy generation

[0119] The decision control unit receives the stain component analysis result, queries the cleaning strategy mapping table, and obtains the basic cleaning mode parameters (such as suction force, water quantity, roller brush rotation speed, etc.).

[0120] 6. Dynamic optimization adjustment

[0121] The dynamic optimization subunit receives real-time state data of the robot (power, water quantity, etc.), performs online adjustment on the basic cleaning parameters through a multi-objective optimization algorithm, and generates a final cleaning mode control instruction set.

[0122] 7. Cleaning instruction execution

[0123] The cleaning execution module receives the control instruction set, drives the stepless speed regulation motor, water pump flow control motor and roller brush torque adjustable motor to execute the corresponding cleaning action.

[0124] 8. Dynamic stain map update

[0125] The navigation positioning module provides the current stain position information, and the dynamic stain map modeling unit binds and records the timestamp of the stain component analysis result and position information of this cleaning, and updates the stain distribution and historical data in the environment map.

[0126] 9. Circular execution and prospective adjustment

[0127] The robot continues to move to the next area and repeats steps 2 to 8. The decision control unit can make prospective adjustments to future cleaning behavior according to historical stain data trends, such as increasing the cleaning intensity of a certain area in advance.

[0128] 10. Task completion and hibernation

[0129] After the cleaning task is completed, the robot returns to the charging base, and the system enters hibernation state, waiting for the next task to start.

[0130] The above is only the preferred embodiment of the present application, and is not intended to limit the present application in any form. Although the present application has been disclosed as above with the preferred embodiment, it is not intended to limit the present application. Any person skilled in the art can make some changes or modifications to the above disclosed technical content without departing from the scope of the technical solution of the present application, and any equivalent embodiments with equivalent changes and modifications are still within the scope of the technical solution of the present application.

Claims

1. A cleaning control system for a robotic vacuum cleaner with adaptive mode adjustment, characterized in that, include: Multimodal sensing module, central processing module, and cleaning execution module; The multimodal sensing module is used to collect multi-dimensional raw data of the robot's environment and ground stains, and includes a hyperspectral imaging unit, an acoustic resonance analysis unit, and a chemical sensing unit. The hyperspectral imaging unit is configured to emit a broadband light beam toward a ground target area and receive reflected light signals to obtain hyperspectral image data of the target area. The acoustic resonance analysis unit is configured to transmit acoustic signals of a specific frequency to the ground target area and receive acoustic echo signals reflected back from the ground target area in order to obtain acoustic response data characterizing the physical properties of the stain. The chemical sensing unit is configured to adsorb and detect gas molecules volatilized from the target area on the ground to obtain gas concentration data characterizing the chemical composition of the stain. The central processing module is electrically connected to the multimodal sensing module and is used to receive and process the multidimensional raw data. It includes a data fusion unit and a decision control unit. The data fusion unit is configured to preprocess, extract features, and perform fusion calculations on the received hyperspectral image data, acoustic response data, and gas concentration data to generate a comprehensive stain composition analysis result, which includes at least a stain type identifier, an adhesion strength quantification value, and a chemical composition identifier. The decision control unit is connected to the data fusion unit and is configured to query a preset cleaning strategy mapping table based on the stain composition analysis results to generate a corresponding cleaning mode control instruction set. The cleaning execution module is electrically connected to the central processing module and is used to receive the cleaning mode control instruction set and execute the cleaning actions defined by the control parameters.

2. The adaptive mode adjustment cleaning control system for a sweeping robot according to claim 1, characterized in that, The data fusion unit includes a hyperspectral feature extraction subunit, an acoustic feature extraction subunit, and a chemical feature extraction subunit; The hyperspectral feature extraction subunit is configured to perform dimensionality reduction and noise reduction processing on the hyperspectral image data, and extract spectral feature vectors that match a known stain sample library. The acoustic feature extraction subunit is configured to perform frequency domain transformation on the acoustic response data to extract energy attenuation features and resonant frequency shift features in a specific frequency band, so as to form an acoustic feature vector. The chemical feature extraction subunit is configured to normalize the gas concentration data and identify the type and concentration ratio of one or more volatile organic compounds based on the sensor response curve to form a chemical feature vector. The data fusion unit also includes a feature fusion subunit based on a deep neural network, which receives the spectral feature vector, acoustic feature vector and chemical feature vector, and performs weighted fusion and joint classification through a trained multi-branch neural network model, and finally outputs the stain component analysis result.

3. The adaptive mode adjustment cleaning control system for a sweeping robot according to claim 2, characterized in that, The feature fusion subunit, based on a multi-branch neural network model using a deep neural network, achieves its final fusion and classification decision through an optimization objective function, which is defined as: in, This indicates the number of sample data in a training batch. This indicates the total number of stain types. It is a binary indicator function that represents a sample. Is the true category? When the true class of the sample is Its value is 1 when it is active, and 0 otherwise. This indicates that the multi-branch neural network model predicts samples. Category The probability, This represents the weight matrix of the hyperspectral feature extraction sub-unit branch network. This represents the weight matrix of the branch network for the acoustic feature extraction subunit. This represents the weight matrix of the chemical feature extraction sub-unit branch network. This represents the L2 norm regularization function applied to the weight matrix. , , These are the regularization hyperparameters corresponding to the weight matrices of each branch network, used to control model complexity and prevent overfitting.

4. The adaptive mode adjustment cleaning control system for a sweeping robot according to claim 2, characterized in that, The decision control unit includes: The cleaning mode control instruction set includes at least the adjustment parameters for the suction motor speed, water tank pump flow rate, roller brush speed, and vibration frequency. Policy mapping subunit and dynamic optimization subunit; The strategy mapping subunit stores the cleaning strategy mapping table, which defines the correspondence between the analysis results of different types of stain components and the basic cleaning mode parameters. The dynamic optimization subunit is configured to receive real-time status data of the robot, including the remaining battery power, the remaining water volume in the clean water tank, and the current capacity of the wastewater tank. The dynamic optimization subunit is further configured to adjust the basic cleaning mode parameters output by the strategy mapping subunit online based on the real-time status data, so as to achieve optimal allocation of cleaning efficiency under system resource constraints, and output the final cleaning mode control instruction set.

5. The adaptive mode adjustment cleaning control system for a sweeping robot according to claim 4, characterized in that, When the dynamic optimization subunit adjusts the basic cleaning mode parameters online based on the real-time status data, it solves the following multi-objective optimization problem: in, Indicates the future starting from the current moment. A sequence of cleaning mode control instructions with a time step. Indicates the length of the optimized time domain. This represents the total energy cost function related to cleaning actions within the optimization time domain. This represents the total water consumption cost function within the optimization time domain. Indicates the first Each time step is determined by the control command. The determined instantaneous power of the system Indicates the first Each time step is determined by the control command. The determined instantaneous flow rate of the water pump and These are the weighting coefficients for energy consumption and water consumption, used to adjust the relative importance of the two optimization objectives. This indicates the current total remaining battery power. This indicates the system's allowed safe battery capacity threshold. This indicates the current total water level remaining in the fresh water tank. This indicates the safe threshold for the amount of clean water allowed by the system. and This indicates the upper and lower limit constraints of the cleaning mode control commands. It represents a time interval of one time step.

6. The adaptive mode adjustment cleaning control system for a sweeping robot according to claim 1, characterized in that, The acoustic resonance analysis unit includes an acoustic transmitter array and an acoustic receiver array. The acoustic wave transmitter array consists of multiple piezoelectric ceramic transducers capable of independently emitting acoustic waves of different frequencies. The acoustic receiver array consists of multiple microphone sensors, used to synchronously receive the acoustic echo signals reflected back from different angles; The acoustic resonance analysis unit controls the acoustic transmitter array to emit frequency scanning signals according to a specific time sequence, and collects the echoes via the acoustic receiver array, thereby calculating the acoustic impedance characteristics and viscoelastic modulus of the stain as an important component of the acoustic response data.

7. The adaptive mode adjustment cleaning control system for a sweeping robot according to claim 1, characterized in that, The chemical sensing unit employs a solid-phase microextraction-gas chromatography-mass spectrometry (SPME-GC-MS) technology module. It includes a miniature sampling pump, a solid-phase microextraction fiber head, a miniature heated desorption chamber, and a metal oxide semiconductor gas sensor array; The micro sampling pump is used to extract gas near the ground and blow it through the solid-phase microextraction fiber head; The solid-phase microextraction fiber head is used to adsorb and enrich volatile organic molecules in the gas. The micro-heating desorption chamber is used to heat the solid-phase microextraction fiber head after adsorption saturation, thereby desorbing the volatile organic molecules. The metal oxide semiconductor gas sensor array is exposed to the desorbed gas, generating electrical signal changes corresponding to specific chemical groups, thereby generating the gas concentration data.

8. The adaptive mode adjustment cleaning control system for a sweeping robot according to claim 1, characterized in that, The central processing module also includes a dynamic stain mapping modeling unit; The dynamic stain mapping modeling unit is communicatively connected to the data fusion unit and the robot's navigation and positioning module; The dynamic stain mapping modeling unit is configured to receive the stain component analysis results from the data fusion unit and the stain geographical location information from the navigation and positioning module; The dynamic stain map modeling unit binds the stain component analysis results with the stain geographical location information and records the timestamp, thereby constructing and updating a dynamic stain map in real time on the robot's environmental map. The dynamic stain map not only marks the spatial distribution of stains, but also stores the historical component analysis results sequence of each stain point in a hierarchical data structure.

9. The adaptive mode adjustment cleaning control system for a sweeping robot according to claim 8, characterized in that, The decision control unit is also connected to the dynamic stain map modeling unit; The decision control unit is configured to, when generating the cleaning mode control instruction set, not only rely on the stain component analysis results generated in real time by the current data fusion unit, but also call the historical component analysis result sequence corresponding to the geographical location in the dynamic stain map; The decision control unit analyzes the changing trends of the historical component analysis result sequence to predict the evolution direction of the stain state and makes forward-looking adjustments to the cleaning mode control instruction set.

10. A cleaning control system for an adaptive mode adjustment sweeping robot according to any one of claims 1-9, characterized in that, The cleaning execution module includes a suction stepless speed-regulating motor, a water pump flow control motor, and a roller brush torque-adjustable motor; The continuously variable speed motor is configured to receive instructions regarding the motor speed from the cleaning mode control instruction set and drive the fan to achieve continuous adjustment of suction power. The water pump flow control motor is configured to receive instructions regarding the water tank pump flow rate from the cleaning mode control instruction set, and drive the micro gear pump to achieve precise control of the water output. The roller brush torque-adjustable motor is configured to receive instructions regarding the roller brush speed and vibration frequency from the cleaning mode control instruction set, and drive the roller brush assembly to operate at a specified speed and torque mode, wherein the vibration frequency is achieved by changing the pulse frequency of the motor input current.

Images