A ground material identification method, system and storage medium of a sweeping robot

Abstract

The application discloses a floor material identification method and system for a sweeping robot and a storage medium. The method synchronously collects a reflection intensity signal of a floor, a driving wheel current signal and a robot motion speed signal, and performs time synchronization and preprocessing on the multiple source signals; a fusion feature is constructed based on the preprocessed signals, and a classification model is used to identify the floor material type. The application adopts reflection intensity normalization and sliding window average processing, and combines a current-speed relationship model to fit a characteristic parameter, effectively overcoming the limitation that a single sensor is easily disturbed, and realizing rapid and accurate identification of various floor materials such as wooden floors, ceramic tiles and carpets under different motion states. The system can be integrated in a sweeping robot, and the suction force is adaptively adjusted according to the identification result, so that the cleaning efficiency and energy efficiency are significantly improved, and the system has strong engineering practicability and popularization value.

Description

Technical Field

[0001] This invention relates to the field of intelligent device technology, specifically to a method, system, and storage medium for identifying floor materials in a robotic vacuum cleaner. Background Technology

[0002] With the rapid development of smart home technology, robotic vacuum cleaners have become increasingly popular as an important tool for household cleaning. Most existing robotic vacuum cleaners possess basic functions such as automatic cleaning, path planning, and automatic recharging. However, the optimal cleaning strategies (such as suction power and movement speed) often differ depending on the floor material (e.g., wood flooring, tile, carpet). For example, stronger suction is needed to effectively remove dust from carpets, while excessive suction on hard, smooth floors not only wastes energy but may also generate unnecessary noise.

[0003] Currently, some high-end robotic vacuum cleaners can automatically adjust their working modes based on the floor material. The key to achieving this function lies in the accurate and rapid identification of the floor material. Existing technologies commonly employ methods for floor material identification, including image recognition based on visual sensors (such as cameras), reflectivity detection based on infrared reflection, or analysis based on motor current load characteristics. However, single-sensor identification methods have significant limitations: for example, visual recognition is easily affected by ambient light; reflectivity detection is prone to misjudgment on surfaces with similar colors but different materials; and relying solely on the drive motor current makes it difficult to distinguish the combined effects of speed changes and load changes on the current.

[0004] Furthermore, existing recognition schemes often lack strict temporal synchronization of multi-sensor signals and do not adequately consider the interference of signal noise and changes in robot motion state on recognition stability. This results in the recognition accuracy and real-time performance of floor materials failing to meet the requirements of adaptive cleaning control during dynamic cleaning processes.

[0005] Therefore, providing a method, system, and storage medium for identifying floor materials in a sweeping robot to solve the above problems is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] To address the aforementioned technical problems, the present invention aims to provide a method for identifying the ground material of a robotic vacuum cleaner. This method can integrate multi-source information, has strong anti-interference capabilities, and can adapt to the dynamic working environment of the robotic vacuum cleaner.

[0007] The technical solution provided by this invention is as follows: A method for identifying the floor material of a robotic vacuum cleaner includes the following steps: Acquire the ground reflection intensity signal; Collect the current signal of the drive wheel and the movement speed signal of the sweeping robot; The reflection intensity signal, the current signal, and the motion speed signal are synchronized based on the same time reference, and preprocessed respectively. Based on the preprocessed reflection intensity signal, preprocessed current signal, and preprocessed motion velocity signal, a fusion feature is constructed, and a classification model is used to identify the ground material type.

[0008] Preferably, the preprocessing of the reflection intensity signal specifically involves: performing normalization and sliding window averaging on the synchronized reflection intensity signal in sequence; The formula for normalization is: ; in, This is the normalized reflection intensity signal. For the reflected intensity signal, and These are the preset maximum and minimum values ​​for reflection intensity.

[0009] Preferably, the preprocessing of the current signal and the motion speed signal includes the following steps: After the synchronized current signal and the synchronized motion speed signal are processed by sliding window averaging, noise suppression is performed by Kalman filtering.

[0010] Preferably, the preprocessing of the current signal and the motion speed signal further includes the following steps: A relationship model is established based on the filtered current signal and the filtered motion velocity signal; Based on the aforementioned relationship model and the least squares method, the first feature parameter and the second feature parameter are obtained through fitting.

[0011] Preferably, the formula for the relational model is: ; in, The synchronized current signal The synchronized motion speed signal and These are the first and second characteristic parameters used to characterize the ground material. These are the model parameters.

[0012] Preferably, it further includes: During the process of identifying the ground material type, if the square of the correlation coefficient between the preprocessed reflection intensity signal and the preprocessed motion speed signal is greater than a preset threshold, the ground material switching mechanism is triggered.

[0013] Preferably, the classification model is a machine learning model trained using a feature database containing samples of various ground materials; The machine learning model is specifically an XGBoost algorithm model or a CNN-LSTM hybrid model; When the CNN-LSTM hybrid model is used, it is used to identify the temporal patterns of the fused features.

[0014] A floor material recognition system for a robotic vacuum cleaner includes: The signal acquisition module is used to acquire the reflected intensity signal from the ground. The signal acquisition module is also used to collect the current signal of the drive wheel and the motion speed signal of the sweeping robot; A time synchronization and preprocessing module is used to synchronize the reflection intensity signal, the current signal, and the motion speed signal based on the same time reference, and to preprocess them respectively. The ground material type identification module is used to construct fused features based on preprocessed reflection intensity signals, preprocessed current signals, and preprocessed motion speed signals, and to identify the ground material type using a classification model.

[0015] A robotic vacuum cleaner integrates a floor material recognition system as described above and adjusts the suction power according to the type of floor material.

[0016] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the floor material recognition method for a sweeping robot as described in any of the preceding claims.

[0017] This invention discloses a method for identifying the ground material of a sweeping robot. The method involves acquiring the reflection intensity signal of the ground through the sweeping robot; collecting the current signal of the sweeping robot's drive wheel and the movement speed signal of the sweeping robot; performing time synchronization on the above-mentioned reflection intensity signal, current signal and movement speed signal based on the same time reference; performing preprocessing on each of the above-mentioned reflection intensity signal, current signal and movement speed signal; constructing a fusion feature based on the preprocessed reflection intensity signal, current signal and movement speed signal; inputting the fusion feature into a classification model; and outputting the classification result to identify the ground material type.

[0018] The present invention has the following beneficial effects: Multi-source information fusion: It comprehensively utilizes optical reflection characteristics and motor load characteristics to describe the ground material from different physical dimensions, overcoming the limitations of a single sensor and significantly improving the recognition accuracy.

[0019] Dynamic characteristic modeling: By establishing a current-velocity relationship model and extracting feature parameters (k1, k2), the influence of robot motion speed on the current signal is effectively removed, making the extracted features more essentially reflect the ground material.

[0020] Robust preprocessing: A series of signal processing techniques, such as normalization, moving average, and Kalman filtering, are employed to effectively suppress noise and improve the system's stability in complex environments.

[0021] Intelligent decision-making mechanism: A material switching triggering mechanism based on correlation coefficients is introduced, which can cope with emergencies, prevent misjudgments, and enhance the system's adaptability.

[0022] The present invention also discloses a ground material recognition system, a sweeping robot, and a computer-readable storage medium. Since these belong to the same technical concept and solve the same technical problem as the method, they should have the same beneficial effects, and will not be described in detail here. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart of a method for identifying the ground material of a sweeping robot provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a floor material recognition system for a sweeping robot provided in an embodiment of the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] The embodiments of this invention are written in a progressive manner.

[0027] This invention provides a method, system, robot, and computer-readable storage medium for identifying floor materials using a robotic vacuum cleaner. It primarily addresses the technical problems in existing technologies, such as susceptibility to misjudgment due to limited information, weak anti-interference capabilities, and incompatibility with the working environment of robotic vacuum cleaners.

[0028] like Figure 1 As shown, a method for identifying the floor material of a robotic vacuum cleaner includes the following steps: S1. Obtain the ground reflection intensity signal; S2. Collect the current signal of the drive wheel and the movement speed signal of the sweeping robot; S3. Time synchronization of the reflection intensity signal, current signal, and motion speed signal based on the same time reference is performed, and preprocessing is performed separately for each signal. S4. Based on the preprocessed reflection intensity signal, preprocessed current signal, and preprocessed motion velocity signal, a fusion feature is constructed, and a classification model is used to identify the ground material type.

[0029] The reflection intensity signal in step S1 refers to the electrical signal obtained by the robotic vacuum cleaner after emitting a beam of light signal through its onboard active optical sensor (usually an infrared sensor, or possibly a LiDAR), and measuring the intensity of the light signal reflected back from the ground. This signal is essentially a time-varying, quantized numerical sequence that directly reflects the optical reflection characteristics of the ground directly below the sensor. The current signal in step S2 refers to the real-time operating current of the drive wheel motor of the sweeping robot. It is a continuous electrical signal that changes over time and directly reflects the load (torque) borne by the motor when driving the wheel. The motion speed signal refers to the physical quantity used to describe the speed and direction of the sweeping robot's movement, usually referring to linear velocity. It is a continuous signal that changes over time and characterizes the robot's real-time motion state during the cleaning process. Time synchronization in step S3 refers to the process of unifying multiple data streams from different sensors (reflection intensity sensor, current sensor, motion speed sensor) onto a common time reference. Its purpose is to ensure that at any given point in time... All the analyzed signal samples (reflection intensity value, current value, velocity value) were indeed collected at the same moment or within the same small time period, reflecting different aspects of the robot in the same state; preprocessing refers to a series of cleaning, enhancement and standardization operations performed on the original signal before using sensor data for feature extraction and model recognition. Its core purpose is to improve signal quality and eliminate or reduce the influence of irrelevant noise and variability. The feature fusion in step S4 refers to the process of combining preprocessed information from multiple independent sensors (reflection intensity, current, and motion speed) into a unified and more discriminative feature vector. This newly constructed feature vector can comprehensively describe the characteristics of the current ground material from multiple physical dimensions, and its recognition capability far surpasses that of any single feature. The classification model is a mathematical model or algorithm whose function is to receive the input vector composed of "fusion features," perform internal calculations and inferences, and output a discrete category label, i.e., the identified ground material type.

[0030] Steps S1 to S4 detail the implementation of the floor material recognition method for the robotic vacuum cleaner, integrating optical characteristics (reflection intensity) and dynamic characteristics (current-velocity relationship). This effectively overcomes the inherent limitations of a single sensor. The preprocessing steps (moving average, Kalman filtering) effectively suppress sensor noise and random disturbances, making the extracted features more stable and reliable, less susceptible to transient outliers. This method models the relationship between current and velocity using a relational model, thus maintaining recognition stability at different movement speeds, which is crucial for robots requiring variable-speed cleaning. Synchronization based on the same time reference ensures that the analyzed data accurately reflects the robot's state at the same moment, avoiding logical errors caused by signal delays and improving the accuracy of real-time decision-making. By monitoring the correlation coefficient between reflection intensity and movement speed and setting a threshold-triggered switching mechanism, a fault-safe and self-checking function is added to the system. When abnormal situations (such as temporary sensor obstruction or extreme ambient light interference) cause unexpected signal correlations, the system can proactively take countermeasures (such as recalculation or entering a conservative mode) to prevent continuous misjudgments, demonstrating the system's intelligence and reliability.

[0031] Specifically, the ground material switching trigger mechanism can perform one or more of the following adaptive behaviors: At the data layer, the above scheme obtains high-quality and consistent raw data through multi-source sensing and time synchronization; at the feature layer, it extracts highly discriminative features with strong anti-interference capabilities and clear physical meaning through advanced preprocessing and feature fusion models; at the decision layer, it achieves highly accurate, robust, and adaptive real-time recognition by utilizing machine learning classification models and intelligent switching mechanisms.

[0032] Preferably, the preprocessing of the reflection intensity signal specifically involves: performing normalization and sliding window averaging on the synchronized reflection intensity signal in sequence; The formula for normalization is: ; in, This is the normalized reflection intensity signal. For the reflected intensity signal, and These are the preset maximum and minimum values ​​for reflection intensity.

[0033] The above details the preprocessing of the reflected intensity signal after time synchronization. A normalization formula maps the absolute intensity value to a relative proportional value, which is fixed within the interval [0, 1]. and Typically preset, representing the sensor's output range under extreme conditions (e.g., Suitable for very dark surfaces or surfaces with no reflection. (Suitable for white ceramic tiles with extremely high reflectivity), regardless of how the absolute reading of the sensor drifts, as long as the relative reflectivity of the ground material remains constant, the normalized value will remain the same. This allows the trained classification model to remain stable. It eliminates reliance on specific hardware units, enhancing the model's generalization ability and the system's long-term stability. The sliding window averaging process uses a fixed-width "window" to slide across the signal and calculates the average of all data within the window as the new signal value at the window's center point. This effectively filters out high-frequency random noise, making the signal curve smoother. With a reasonable window width, it can filter out noise while preserving the low-frequency main trends of ground material changes. For example, the overall decreasing trend of reflectivity from wood flooring to carpet is clearly preserved, while small fluctuations in the process are smoothed out. Through averaging, random noise components in the signal cancel each other out, while information representing the true material is enhanced.

[0034] The above scheme addresses the consistency and generalization issues of the system through normalization, and solves the noise and stability problems of the signal through sliding window averaging. The combination of these two methods significantly improves the quality of the reflection intensity signal with extremely low computational cost, laying a solid foundation for building high-performance fusion feature and classification models.

[0035] In one embodiment, the specific implementation process of the sliding window averaging of the reflection intensity signal is as follows: 1. Input parameters Original reflection intensity sequence:

[0036] Window width: (Typical value 5-15, adjust according to sampling frequency) Sliding step size: = 1 (Default: point-by-point sliding) 2. Processing Algorithm For each output position (from 1 to ): Output value [ ] = Average ( arrive ) 3. Specific examples Original signal: [0.85, 0.83, 0.82, 0.25, 0.23, 0.24] (Floor → Carpet switch) Window width: = 3 Processing procedure: Window 1: (0.85 + 0.83 + 0.82) / 3 = 0.833 Window 2: (0.83 + 0.82 + 0.25) / 3 = 0.633 Window 3: (0.82 + 0.25 + 0.23) / 3 = 0.433 Window 4: (0.25 + 0.23 + 0.24) / 3 = 0.240 4. Technical Effects Noise suppression: Smooths out random fluctuations and transient interference from the sensor; Trend preservation: Maintains the overall trend of material changes but makes it smoother; Boundary transition: Creates gradual transitions at material boundaries rather than abrupt jumps; 5. Project Implementation Real-time calculation: Sliding and optimization are used to avoid redundant calculations; Boundary handling: Cumulative averaging or zero-filling can be used in the initial stage; Parameter tuning: The window size is optimized according to the robot's movement speed and sampling rate.

[0037] Preferably, the preprocessing of the current signal and the motion velocity signal includes the following steps: After the synchronized current signal and the synchronized motion speed signal are processed by sliding window averaging, noise suppression is performed by Kalman filtering.

[0038] Kalman filtering is an optimal recursive digital signal processing algorithm used to estimate the internal true state of a dynamic system from a series of noisy observation data. Its core idea is to combine "prediction" (based on the system model) and "update" (based on sensor observations) through probability weighting to obtain a more accurate and smoother state estimate than a single prediction or observation. The above scheme preprocesses the current signal and motion speed signal after time synchronization. By using a sliding window averaging process as a pre-filter, it primarily smooths and initially screens the signal, suppressing high-frequency impulse noise. The current signal is highly susceptible to brief spikes caused by the impact of wheels momentarily running over small stones, electrical wires, or floor seams. The sliding average can quickly and effectively smooth out these sudden, short-lived strong interferences, preventing them from directly entering the Kalman filter and avoiding the filter being skewed by these outliers; it also reduces the computational load. While Kalman filtering is accurate, it is computationally expensive. Using a computationally simple sliding average for a preliminary "coarse processing" can smooth out most of the high-frequency noise beforehand, reducing the processing burden on the subsequent Kalman filter and making the entire system more suitable for real-time operation on the embedded processor of the robotic vacuum cleaner; it also provides an initial smoothing trend. It provides the Kalman filter with a relatively clean and clearly defined initial input, helping the filter converge to the true state more quickly. Using Kalman filtering as the core fine filter, optimal estimation is achieved through the system model: optimal recursive noise suppression. Kalman filtering goes beyond mere smoothing; it models the dynamic behavior of the system (such as the robot's inertia and motor response characteristics) through a state-space model, intelligently distinguishing the true trend of the signal from random noise and providing the statistically optimal estimate (minimum mean square error). Its filtering effect is smoother and more accurate than simple moving average; dynamic tracking and prediction capabilities are also provided. When the robot accelerates, decelerates, or goes up or down slopes, the current and velocity undergo expected, dynamically consistent changes. The Kalman filter can predict these changes and distinguish them from abnormal fluctuations caused by changes in ground material; it provides the confidence level (covariance) of the state estimate. The Kalman filter not only outputs the estimate but also its uncertainty (covariance). This information can be used by subsequent systems.

[0039] The above scheme combines real-time smoothing by sliding window averaging with optimal estimation by Kalman filtering to form an efficient and robust signal processing pipeline, ultimately providing the system with current and velocity information that can accurately reflect the characteristics of the ground load.

[0040] Preferably, the preprocessing of the current signal and the motion speed signal further includes the following steps: A relationship model is established based on the filtered current signal and the filtered motion velocity signal; Based on the relational model and the least squares method, the first feature parameter and the second feature parameter are obtained by fitting.

[0041] Preferably, the formula for the relational model is: ; in, The synchronized current signal The synchronized motion speed signal and These are the first and second characteristic parameters used to characterize the ground material. These are the model parameters.

[0042] The above steps involve further processing the filtered current signal and the filtered velocity signal after Kalman filtering. Specifically, this is done by fitting a data segment using the least squares method. and What we get is no longer a current signal that varies with time, but two constants that characterize the overall drag characteristics of the ground during that period. This applies regardless of how the velocity changes within this data range (acceleration, deceleration, or constant speed). and Both remain relatively stable, achieving perfect decoupling between speed and load; (First-order coefficient): This can be understood as a resistance term proportional to speed, and is usually strongly correlated with the rolling friction characteristics of the ground. The k1 value of carpet is significantly higher than that of ceramic tile; (Second-order coefficients): These can be understood as drag terms proportional to the square of the velocity, and in some cases can also capture nonlinear load effects; due to and These data are fitted from a time-series dataset, making them insensitive to transient noise in the signal. Errors at individual data points are averaged out during the least-squares fitting process. This method makes the system independent of any specific velocity point. Whether the robot is performing quiet, slow cleaning or rapid area traversal, as long as sufficient data segments are collected, a model reflecting the current ground material can be fitted. and This greatly enhances the recognition system's ability to generalize across different operating modes.

[0043] The above scheme uses physical models and mathematical fitting to extract the parts related to motion state from dynamically changing current and velocity signals, and extracts robust feature parameters that are only related to the ground material. It has the following advantages: the features directly correspond to the physical essence, avoiding confusion caused by velocity changes; it is not sensitive to noise and velocity changes; and the input features have low dimensionality and clear meaning.

[0044] Preferably, it further includes: During the process of identifying the ground material type, if the square of the correlation coefficient between the preprocessed reflection intensity signal and the preprocessed motion speed signal is greater than a preset threshold, the ground material switching mechanism is triggered.

[0045] The above scheme introduces a correlation coefficient between reflection intensity and movement speed, targeting the following scenarios: Sensor malfunction or contamination: The reflection intensity sensor is partially obscured by dust or hair, resulting in generally low readings. If the robot happens to be accelerating (e.g., leaving an obstacle), the sensor moves away from the obstruction, and the reading may momentarily rise, creating a false correlation with speed changes. Special environmental interference: When the robot enters an extremely dark environment (e.g., under a bed or cabinet), the reflection intensity signal remains low, and the relative noise level increases. In this case, any small change in speed may create a non-causal statistical correlation with the noise fluctuations in the signal. By setting the aforementioned correlation coefficient and preset threshold, illogical signal relationships are detected and used as an indicator of unreliable system status. This indicator adds a credibility verification dimension to the main classification model's decision, making the system no longer "blind" but possessing a certain degree of "self-doubt" and "self-checking" capabilities. The triggering ground material switching mechanism can be specified into the following intelligent behaviors: Decision reset: Ignore the current classification result based on the anomalous signal and retain the last material state that was reliably identified; Conservative mode switching: Switch to a preset, safe, and universal cleaning mode (such as medium suction) to avoid performing incorrect operations based on misjudgment (such as turning on the maximum suction on a carpet that is misjudged, thus wasting power). Trigger recalibration: Initiate a more comprehensive signal detection or sensor self-test process to attempt to restore the system to normal operation. Increase the decision threshold: Temporarily increase the confidence threshold required for the classification model to make a decision, and only output a result when it is very certain, otherwise remain silent.

[0046] The above solution upgrades the system from an open-loop, passively classified system to a closed-loop intelligent system with perception, decision-making, status monitoring, and proactive response capabilities. By monitoring the abnormal correlation between two physical quantities that should be independent—reflection intensity and movement speed—it can capture systemic anomalies that are difficult for the main model to detect. Through a trigger switching mechanism, it achieves fault mitigation, ensuring that the entire robot vacuum cleaner can operate stably and reliably in complex and unstructured home environments.

[0047] Preferably, the classification model is a machine learning model trained using a feature database containing samples of various ground materials; The machine learning model is specifically the XGBoost algorithm model or the CNN-LSTM hybrid model; When a CNN-LSTM hybrid model is used, it is employed to identify temporal patterns of fused features.

[0048] The above solutions provide optimal model selection for different data characteristics and application needs, ensuring both high-precision static classification capabilities and advanced capabilities to capture dynamic temporal patterns, thereby comprehensively improving the accuracy and environmental adaptability of ground material identification. The solution offers a complete technology spectrum from "lightweight and efficient" (XGBoost) to "heavyweight and high-precision" (CNN-LSTM), allowing for flexible selection based on product positioning and hardware costs.

[0049] Performance versus cost trade-offs: XGBoost: Represents the engineering-optimal solution for achieving the best performance with limited computing resources.

[0050] CNN-LSTM represents a cutting-edge solution that pursues ultimate recognition performance and spares no expense in computing resources.

[0051] The evolution of feature engineering: XGBoost relies on the quality of human feature engineering, embodying the wisdom of traditional machine learning.

[0052] CNN-LSTM enables automatic feature learning, reducing reliance on manually designed features and representing the future direction of deep learning.

[0053] The above scheme ensures that the entire ground material recognition system can provide a stable, efficient, and feasible benchmark solution through the XGBoost model, and can also explore the performance limit of temporal recognition through the CNN-LSTM hybrid model, reserving technical space for achieving more intelligent and human-like environmental perception capabilities.

[0054] In one embodiment, steps S1 to S4 are specifically implemented as follows: The scenario is set as follows: the robot vacuum is working normally in the home environment, moving from the hardwood floor area to the short-pile carpet area. 1. Multi-source signal acquisition and time synchronization Signal acquisition: Reflection intensity signal The robot's downward-facing sensor (LiDAR / ToF module) emits infrared light at a frequency of 100Hz and receives the ground reflection intensity signal. The raw signal is a sequence of voltage values ​​that change over time.

[0055] Current signal The drive motor controller collects the motor's operating current at a frequency of 100Hz.

[0056] Motion speed signal The wheel encoder calculates the robot's real-time linear velocity at a frequency of 100Hz.

[0057] Time synchronization: The system uses a high-precision hardware clock as its time base, assigning a unique timestamp to each collected data sample. .

[0058] By using interpolation or data alignment algorithms, it is ensured that at any given moment in processing, the data used for analysis is accurate and consistent. All data segments strictly correspond to the same time period.

[0059] 2. Signal preprocessing Reflection intensity signal preprocessing Normalization: Preset = 3.3V (corresponding to pure white high-reflectivity ceramic tile), = 0.1V (corresponding to extremely dark carpet or black hole).

[0060] Original reflection intensity signal Substitute into the formula This maps all values ​​to the interval [0, 1].

[0061] Sliding window averaging: Set window width = 10 (corresponding to a 100ms time window).

[0062] For the normalized sequence Perform a point-by-point moving average.

[0063] Example: Before processing = [0.85, 0.83, 0.82, 0.25, 0.23, ...] (floor to carpet), after processing becomes [0.833, 0.633, 0.433, 0.240, ...], the signal becomes smoother, but the overall trend from high to low is preserved.

[0064] Current and velocity signal preprocessing Primary smoothing: right and Similarly, a sliding average with a window width of 10 is applied to initially suppress current spikes caused by overcurrent through wires, seams, etc.

[0065] Kalman filtering: Prediction: Based on the robot's motion state (velocity, acceleration) at the previous moment, predict the current and velocity values ​​at the current moment.

[0066] Update: The predicted value and the sensor measured value are weighted and fused (the weighting coefficient is determined by the Kalman gain, which is dynamically adjusted) to output the optimal estimate.

[0067] Effect: Obtain two extremely smooth lines that faithfully adhere to the laws of physics. and The curve effectively filters out random noise and can intelligently track the robot's acceleration / deceleration process.

[0068] 3. Feature Fusion Construction Extracting dynamic characteristic parameters: Modeling: Take a 200ms segment of filtered data (20 data points) and establish a relational model: ; Fitting: The least squares method was used to perform curve fitting on this data segment to accurately calculate the characteristic parameters of the current ground. and .

[0069] Physical meaning: The value is significantly higher on carpets than on wooden floors because it reflects the basic coefficient of rolling friction of the surface.

[0070] Constructing feature vectors: The following features are combined into a single feature vector, which is then used as input to the classification model: The average value of the preprocessed reflection intensity signal within the same time period.

[0071] The standard deviation of the pre-processed reflection intensity signal within the same time period (reflecting surface uniformity).

[0072] : First-order characteristic parameters obtained from fitting the current-velocity model.

[0073] : Second-order characteristic parameters obtained from fitting the current-velocity model.

[0074] The average value of the filtered velocity signal within the same time period.

[0075] 4. Classification and Intelligent Decision Making Model recognition: The constructed feature vectors are input into the pre-trained XGBoost model.

[0076] The model outputs a probability distribution based on the learned decision rules, for example: [wood flooring: 0.02, tile: 0.01, short-pile carpet: 0.97].

[0077] The system determines that the current floor is a short-pile carpet.

[0078] Redundancy checks and security mechanisms: Continuous monitoring: Throughout the process, the system calculates the preprocessed reflection intensity signal in real time. With speed signal The square of the correlation coefficient (R²).

[0079] Triggering condition: When the robot crawls under the bed, the sensors are briefly contaminated by dust, causing... Abnormal fluctuations and Create a strong association, making (Preset threshold)

[0080] Intelligent response: The system immediately triggers the material switching mechanism and adopts "conservative mode": ignores the current XGBoost output, maintains the suction power at a medium level, records the area that needs to be cleaned, and re-identifies after the signal returns to normal.

[0081] 5. Implementation and Feedback When the system reliably identifies the short-pile carpet with extremely high confidence, it sends a command to the main controller.

[0082] The main controller then adjusts the fan to increase suction to carpet cleaning mode, ensuring that deep dust is effectively removed.

[0083] At the same time, the area is marked with material labels in the machine's memory map to facilitate targeted path planning in the future.

[0084] In this embodiment, the robot successfully achieved seamless and accurate identification from wooden floors to carpets, and adaptively adjusted suction power based on the identification results. The entire process overcomes the shortcomings of single sensors being susceptible to interference. Through multi-source fusion and intelligent preprocessing, it maintains high robustness in dynamic environments, ultimately achieving an energy-saving, efficient, and intelligent cleaning experience.

[0085] like Figure 2 As shown, a floor material recognition system for a robotic vacuum cleaner includes: The signal acquisition module is used to acquire the reflected intensity signal from the ground. The signal acquisition module is also used to collect the current signal of the drive wheel and the movement speed signal of the sweeping robot; The time synchronization and preprocessing module is used to synchronize the reflection intensity signal, current signal and motion speed signal based on the same time reference, and to perform preprocessing on each signal separately. The ground material type identification module is used to construct fused features based on preprocessed reflection intensity signals, preprocessed current signals, and preprocessed motion speed signals, and to identify the ground material type using a classification model.

[0086] The present invention also discloses a floor material recognition system for a sweeping robot, which aims to achieve floor material recognition based on the sweeping robot in a modular manner.

[0087] A robotic vacuum cleaner integrates a floor material recognition system as described above and adjusts the suction power according to the type of floor material.

[0088] The present invention also discloses a sweeping robot that deeply integrates high-precision ground material recognition capability with suction adjustment function, realizing full-link intelligence from "perceiving the environment" to "intelligent decision-making" to "precise execution", ultimately achieving a comprehensive improvement in cleaning effect, energy efficiency and user experience.

[0089] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a floor material recognition method for a sweeping robot as described above.

[0090] The present invention also discloses a computer-readable storage medium that transforms the aforementioned innovative and complex multi-source sensor fusion identification method from a technical concept into a standardized, replicable, and deployable industrial product.

[0091] One or more embodiments in this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of this application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments in this application should be included within the protection scope of this application.

[0092] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature.

[0093] If a flowchart is used in this application, it is used to illustrate the operations performed by the system according to embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, the steps can be processed in reverse order or simultaneously. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.

[0094] The foregoing has provided a detailed description of a method, system, robot, and storage medium for identifying floor materials in a robotic vacuum cleaner, as provided in this application. The above description of the disclosed embodiments enables those skilled in the art to implement or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for identifying the floor material of a robotic vacuum cleaner, characterized in that, Includes the following steps: Acquire the ground reflection intensity signal; Collect the current signal of the drive wheel and the movement speed signal of the sweeping robot; The reflection intensity signal, the current signal, and the motion speed signal are synchronized based on the same time reference, and preprocessed respectively. The preprocessing of the reflection intensity signal specifically involves: performing normalization and sliding window averaging on the synchronized reflection intensity signal in sequence; The formula for normalization is: ； in, This is the normalized reflection intensity signal. For the reflected intensity signal, and These are the preset maximum and minimum values ​​for reflection intensity; The preprocessing of the current signal and the motion velocity signal includes the following steps: After the synchronized current signal and the synchronized motion speed signal are processed by sliding window averaging, noise suppression is performed by Kalman filtering. Preprocessing the current signal and the motion speed signal further includes the following steps: A relationship model is established based on the filtered current signal and the filtered motion velocity signal; Based on the aforementioned relationship model and the least squares method, the first feature parameter and the second feature parameter are obtained by fitting. The formula for the relational model is: ； in, The synchronized current signal The synchronized motion speed signal and These are the first and second characteristic parameters used to characterize the ground material. These are model parameters; Based on the preprocessed reflection intensity signal, preprocessed current signal, and preprocessed motion velocity signal, a fusion feature is constructed, and a classification model is used to identify the ground material type. The steps to construct the fusion feature are as follows: combine the following features into a single feature vector, which will serve as the fusion feature: The average value of the preprocessed reflection intensity signal within the same time period. The standard deviation of the preprocessed reflection intensity signal within the same time period. The first characteristic parameter obtained from fitting the current-velocity model. The second characteristic parameter obtained by fitting from the current-velocity model. The average value of the filtered velocity signal within the same time period.

2. The method for identifying the floor material of a sweeping robot as described in claim 1, characterized in that, Also includes: During the process of identifying the ground material type, if the square of the correlation coefficient between the preprocessed reflection intensity signal and the preprocessed motion speed signal is greater than a preset threshold, the ground material switching mechanism is triggered.

3. The method for identifying the floor material of a sweeping robot as described in claim 1, characterized in that, The classification model is a machine learning model trained using a feature database containing samples of various ground materials. The machine learning model is specifically an XGBoost algorithm model or a CNN-LSTM hybrid model; When the CNN-LSTM hybrid model is used, it is used to identify the temporal patterns of the fused features.

4. A floor material recognition system for a robotic vacuum cleaner, characterized in that, include: The signal acquisition module is used to acquire the reflected intensity signal from the ground. The signal acquisition module is also used to collect the current signal of the drive wheel and the motion speed signal of the sweeping robot; A time synchronization and preprocessing module is used to synchronize the reflection intensity signal, the current signal, and the motion speed signal based on the same time reference, and to preprocess them respectively. The preprocessing of the reflection intensity signal specifically involves: performing normalization and sliding window averaging on the synchronized reflection intensity signal in sequence; The formula for normalization is: ； in, This is the normalized reflection intensity signal. For the reflected intensity signal, and These are the preset maximum and minimum values ​​for reflection intensity; The preprocessing of the current signal and the motion velocity signal includes the following steps: After the synchronized current signal and the synchronized motion speed signal are processed by sliding window averaging, noise suppression is performed by Kalman filtering. Preprocessing the current signal and the motion speed signal further includes the following steps: A relationship model is established based on the filtered current signal and the filtered motion velocity signal; Based on the aforementioned relationship model and the least squares method, the first feature parameter and the second feature parameter are obtained by fitting. The formula for the relational model is: ； in, The synchronized current signal The synchronized motion speed signal and These are the first and second characteristic parameters used to characterize the ground material. These are model parameters; The ground material type identification module is used to construct fused features based on preprocessed reflection intensity signals, preprocessed current signals, and preprocessed motion velocity signals, and to identify the ground material type using a classification model; The steps to construct the fusion feature are as follows: combine the following features into a single feature vector, which will serve as the fusion feature: The average value of the preprocessed reflection intensity signal within the same time period. The standard deviation of the preprocessed reflection intensity signal within the same time period. The first characteristic parameter obtained from fitting the current-velocity model. The second characteristic parameter obtained by fitting from the current-velocity model. The average value of the filtered velocity signal within the same time period.

5. A robotic vacuum cleaner, integrating the floor material recognition system as described in claim 4, and adjusting the suction power according to the type of floor material.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method for identifying the floor material of the sweeping robot as described in any one of claims 1 to 3.

Images