A method for controlling the spraying of mite remover by a bed cleaning robot based on path backtracking

Abstract

This invention discloses a method for controlling the spraying of mite-removing agents by a bed cleaning robot based on path backtracking, relating to the field of smart home cleaning equipment technology. The method includes: Step S1, performing initial cleaning and constructing an information-enhanced path: controlling the cleaning robot to perform the initial cleaning operation on the bed surface according to a first planned path, and simultaneously collecting multimodal state information of the bed surface, and binding the multimodal state information to the coordinate points on the first planned path. This invention upgrades the initial cleaning path to an information-enhanced path. The robot not only completes physical cleaning in the initial operation, but also simultaneously collects and binds the bed surface state information of each path point in real time through multimodal sensors, including visual dirt index, allergen concentration, and fabric fluffiness. In the subsequent backtracking spraying stage, the system does not simply repeat the path, but adjusts the spraying parameters dynamically and in real time according to the specific information bound to each location.

Description

Technical Field

[0001] This invention relates to the field of smart home cleaning equipment technology, specifically to a method for controlling the spraying of mite remover by a bed cleaning robot based on path tracing. Background Technology

[0002] With the increasing popularity of healthy living concepts, the demand for deep cleaning and mite removal of bedding is growing. As an automated solution, bed cleaning robots have evolved from simple vacuuming to comprehensive mite removal platforms integrating ultraviolet sterilization, beating, hot air dehumidification, and even chemical spraying. A typical workflow usually includes two main stages: first, a physical cleaning stage to remove dust, dander, and some mites; followed by a chemical treatment stage, which involves spraying mite-removing agents to achieve deep sterilization and inhibit mite regeneration.

[0003] However, in the crucial post-treatment acaricide spraying control stage, existing technologies have significant defects and limitations, mainly in the following three aspects:

[0004] First, the cleaning and spraying processes are disconnected, resulting in low levels of intelligence in spraying control. Existing bed cleaning robots typically treat cleaning and spraying as two independent modules executed in a fixed sequence. Their spraying control mostly employs simple timed spraying or uniform spraying along a fixed path. That is, the robot first completes a full bed cleaning, then strictly or roughly follows the previous trajectory for a second pass and simultaneously starts spraying. This mode usually fails to fully utilize the valuable information obtained during the initial cleaning process regarding the degree of dirt and microenvironment characteristics of different areas of the bed. For example, the concentration of dust and allergens in the central area of ​​the bed and the area where people usually lie may be much higher than in the edge areas; the agent adhesion requirements in the folds of sheets and the fluffy areas of bedding are also significantly different from those in flat areas. Existing technology has limitations in achieving precise, on-demand application of agents, which may lead to insufficient spraying in highly polluted areas, affecting the mite-removal effect, or overspraying in low-polluted areas, resulting in agent waste and potential residues.

[0005] Secondly, the spraying strategy is rigid and unable to adapt to the complex and ever-changing microenvironment of the bedding surface. The bedding surface is not a homogeneous plane; its material, texture, wrinkle depth, and loft vary. These microenvironmental factors directly affect the atomization, penetration, diffusion, and adhesion of the pesticide. Existing spraying systems generally operate with fixed parameters, making it difficult to dynamically adjust based on real-time environmental characteristics. For example, for fluffy bedding, finer atomization and a higher nozzle angle are needed to ensure the pesticide penetrates deep into the fibers; for the edges of the bedding, the spray angle needs to be adjusted to prevent the pesticide from spraying off the edge. The adaptability of existing technologies in this scenario needs improvement, resulting in uneven overall spraying effects, poor reliability, and poor adaptability in complex bedding environments. Summary of the Invention

[0006] This invention addresses the problem of overly simplistic solutions in existing technologies by providing a significantly different approach. Specifically, the invention aims to provide a path-backtracking-based method for controlling the spraying of a mite remover in a bed cleaning robot. This method solves the technical problems of low intelligence in the post-spraying process, inability to accurately adapt to the actual conditions of the bed, and lack of self-optimization and collaborative evolution capabilities in existing technologies.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a method for controlling the spraying of mite remover by a bed surface cleaning robot based on path backtracking, comprising:

[0008] Step S1, Perform initial cleaning and construct information-enhanced path: Control the cleaning robot to perform initial cleaning on the bed surface according to the first planned path, and simultaneously collect multimodal state information of the bed surface, and bind the multimodal state information with the coordinate points on the first planned path to form an information-enhanced path;

[0009] Step S2, Dynamic backtracking spraying based on information-enhanced path: After the first cleaning operation is completed, the cleaning robot is controlled to move along the backtracking path associated with the first planned path. When it moves to the coordinate point on the backtracking path, the multimodal state information corresponding to the coordinate point bound in step S1 is read, and the spraying action of the rear mite spraying component is dynamically adjusted based on the information.

[0010] Preferably, the multimodal state information includes at least two types of heterogeneous information extracted from the bed surface environment:

[0011] The first category is dirt and grime information that characterizes the cleanliness requirements of the bed surface;

[0012] The second category is microenvironmental information that characterizes the impact of the physical morphology of the spraying bed on spraying operations.

[0013] Preferably, the collection of dirt information includes:

[0014] Images of the fabric surface are acquired by a near-field vision sensor placed near the cleaning component, and the visual dirt index is obtained through image processing.

[0015] And / or, the dust concentration value of the inhaled gas is acquired in real time by a particulate sensor integrated in the dust suction channel, wherein the particulate sensor may be a laser scattering particulate sensor.

[0016] Preferably, the collection of microenvironment information includes:

[0017] The fluffiness or stiffness of the bed fabric can be indirectly assessed by pressure sensors or drive motor current feedback.

[0018] And / or, by using the robot's posture sensors in conjunction with path information, identify depressions, wrinkles, or edge areas on the bed surface.

[0019] Preferably, the dynamic adjustment of the spraying action of the post-acaricide spraying component based on this information includes:

[0020] For coordinate point areas that are bound to different levels of dirt information, the spraying dosage is controlled differently;

[0021] For coordinate point areas that are bound to different microenvironment information types, the atomization particle size, spray angle, or nozzle height of the spray can be controlled differently.

[0022] Preferably, step S3 is also included:

[0023] In the dynamic retrograde spraying step, a high-sensitivity metal oxide semiconductor VOC sensor located behind the spraying assembly collects the gas concentration signal after spraying. This sensor is calibrated for common volatile organic components in acaricides. By monitoring the peak value or time integral value of the VOC concentration within a specific time window after spraying, an immediate effect feedback signal reflecting the adhesion of the acaricide to the bed surface is collected.

[0024] The real-time effect feedback signal is associated and stored with the current coordinate point, its bound multimodal state information, and the spraying action performed.

[0025] Preferably, step S4 is also included:

[0026] Based on the accumulated associated data in step S3, the decision mapping relationship from multimodal state information to spraying actions is periodically optimized and updated using machine learning algorithms, so that the spraying actions tend to be better based on historical effect feedback.

[0027] Preferably, step S4 further includes cross-cycle strategy iteration and cleaning-spraying synergistic optimization, specifically including:

[0028] The optimized and updated decision mapping relationship is applied to the dynamic backtracking spraying step when the robot executes the method of claim 1 again;

[0029] And / or, based on specific microenvironment area types identified in historical data where the spraying effect feedback is consistently below a preset threshold, when the robot performs its first cleaning and information binding step again, the robot adaptively adjusts the moving speed, cleaning intensity, or local coverage density of the first planned path for performing the first cleaning operation in such areas.

[0030] Preferably, step S1 further includes:

[0031] When the edge of the bed is detected by the cliff sensor, the robot is controlled to perform contour cleaning along the edge of the bed, and when constructing the information-enhanced path, special boundary area markers are bound to the coordinate points near the edge of the bed.

[0032] In step S2, when tracing back to the coordinate point bound by the boundary area marker, the spraying assembly is controlled to reduce the spray angle or close the spray nozzles located on the outside of the bed edge to prevent the acaricide from being sprayed outside the bed.

[0033] Preferably, the first planned path is a bow-shaped path covering the bed surface, and the backtracking path is a guiding path that partially or completely overlaps with the bow-shaped path after sampling or simplification.

[0034] Compared with the prior art, the beneficial effects of the present invention are:

[0035] 1. This invention achieves a qualitative leap from uniform blind spraying to information-driven, on-demand precision spraying, significantly improving mite removal efficiency and saving on pesticides. This invention creatively upgrades the initial cleaning path to an information-enhanced path. The robot not only completes physical cleaning during the initial operation but also simultaneously collects and binds real-time bed surface status information at each path point through multimodal sensors, including visual dirt index, allergen concentration, and fabric fluffiness. In the subsequent back-spraying phase, the system does not simply repeat the path but dynamically adjusts the spraying parameters in real-time based on the specific information bound to each location. For example, in areas with high dust concentration, it automatically increases the spraying dosage and duration to ensure sufficient treatment of heavily contaminated areas; in identified areas with deep wrinkles or fluffy bedding, it automatically switches to a fine atomization mode and adjusts the nozzle posture to promote pesticide penetration. This real-time closed loop of perception-decision-execution allows pesticide resources to be precisely delivered to the most needed locations and act in the most appropriate way, fundamentally overcoming the drawbacks of traditional uniform spraying.

[0036] 2. This invention constructs an evaluation-optimization self-learning closed loop, enabling the spraying system to possess adaptive and continuously evolving capabilities, significantly enhancing its intelligence. This invention overcomes the limitations of traditional open-loop control systems by introducing an effect feedback and strategy optimization mechanism. By integrating VOC or fluorescent tracer sensors behind the spray head, the system can evaluate the immediate adhesion effect of the agent on the bed surface online, and associate and store this feedback signal with the corresponding location, status information, and spraying actions taken, forming a historical experience database. Periodically, the machine learning module within the system analyzes this data, automatically discovering which spraying strategy yields better results under different bed surface conditions, thereby continuously optimizing the decision mapping rules between status information and spraying actions. This means that the robot's spraying strategy is no longer fixed, but can self-learn and self-adjust with increased usage, gradually adapting to the characteristics and contamination patterns of specific bedding. In the long term, the system can evolve personalized optimal spraying solutions to cope with scenarios such as bedding changes and seasonal variations, achieving a leap from executing fixed programs to possessing learning intelligence, greatly enhancing the long-term use value of the product and the user experience.

[0037] 3. This invention achieves deep coupling and global performance optimization of the two core functions of cleaning and spraying through a cross-cycle co-evolution mechanism. The ingenuity of this invention is further reflected in the fact that its optimization mechanism not only acts on the spraying decision itself but also empowers the initial cleaning process. By analyzing historical data, if the system identifies certain types of microenvironment areas where spraying effects are consistently poor, optimization instructions can be fed back to the path planning and cleaning control modules. In the next operation, the robot can automatically adjust its initial cleaning strategy in such areas, such as reducing movement speed, increasing suction force, or using a denser path for coverage, in order to change the initial state of the area and create more favorable conditions for subsequent precise spraying. This cross-cycle, cross-module co-evolution mechanism—spraying effect feedback—optimized cleaning strategy—improved state information—further enhanced spraying effect—integrates cleaning and spraying from two isolated processes into an organic, mutually reinforcing whole system. Attached Figure Description

[0038] Figure 1 This is a schematic diagram of the process structure of a method for controlling the spraying of mite-removing agent by a bed surface cleaning robot based on path backtracking, according to the present invention. Detailed Implementation

[0039] 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.

[0040] Please see Figure 1 This invention provides a technical solution: a method for controlling the spraying of mite-removing agents by a bed surface cleaning robot based on path backtracking. The core of this method lies in upgrading the traditional geometric movement path to an information-enhanced path carrying rich environmental information, and achieving closed-loop intelligent control of the spraying operation based on this path. The method includes the following sequentially connected or iteratively linked steps:

[0041] First, an initial cleaning is performed, and an information-enhanced path is constructed. The cleaning robot is controlled to perform the initial cleaning operation on the bed surface according to the first planned path. This cleaning operation typically includes physical cleaning actions such as vacuuming and patting. During this process, the robot does not simply move, but acts as a mobile sensing platform, synchronously and in real-time collecting multi-dimensional state information of the bed surface through its onboard multimodal sensor array. This information is not stored in isolation, but is precisely bound to the high-precision spatiotemporal coordinates provided by the robot's positioning system, such as encoders, inertial measurement units (IMUs), and visual odometry—that is, to the various path points on the first planned path. This binding forms an information-enhanced path that goes far beyond a geometric trajectory; it establishes a digital twin model of the bed surface, including the degree of dirt in different areas and the characteristics of the microenvironment in those areas. The first planned path is preferably a highly efficient, full-coverage path, such as a bow-shaped path, to ensure that no information is missed during collection.

[0042] Secondly, dynamic backtracking spraying is performed based on information-enhanced paths. After the initial cleaning operation, the robot does not immediately stop working but begins the second phase of the task. It moves along a backtracking path closely related to the first planned path, which can be a simplification, sampling, or complete reproduction of the first planned path. Crucially, as the robot moves to each coordinate point on the backtracking path, the control system reads in real time the multimodal state information bound to that point and collected during the initial cleaning. Based on this specific information reflecting the unique needs of that point, the controller dynamically and differentially adjusts one or more spraying control parameters of the post-mite spraying component, such as spray flow rate, atomization particle size, nozzle oscillation angle, or height, thereby performing precise spraying operations on demand. This completely changes the extensive mode of spraying uniformly regardless of location in existing technologies.

[0043] Furthermore, the composition of the multimodal state information is based on the extraction of key attributes of the bed surface environment in the inventive concept. This invention extracts two types of heterogeneous information crucial for spraying decisions from the complex bed surface environment: the first type is dirt information directly characterizing the required cleanliness of the bed surface, reflecting where more treatment is needed; the second type is microenvironmental information characterizing the impact of the bed surface's physical morphology on spraying operations, reflecting how treatment should be carried out. The combination of these two types of information provides comprehensive and targeted input for spraying decisions.

[0044] Specifically, the collection of dirt information can be achieved through various sensing methods. One effective approach is to acquire microscopic images of the fabric surface using near-field vision sensors, such as miniature CMOS cameras, placed near the cleaning roller brush or vacuum inlet. By processing and analyzing these images in real time, for example, extracting texture changes and spot features, a visual dirt index can be calculated, which effectively reflects the distribution of visible dander and stains. Simultaneously, to detect invisible allergens, a highly sensitive laser particle sensor can be integrated into the robot's vacuum duct to monitor the concentration of dust and allergen particles in the inhaled airflow in real time. The fusion of visual and particle information allows for a comprehensive assessment of the bed surface's contamination load from both visible and invisible dimensions.

[0045] The collection of microenvironment information focuses on the physical characteristics of the bed surface. For example, by monitoring changes in the current load of the drive wheel motor or by using an additional pressure sensor array on the chassis, the fluffiness or stiffness of the bed fabric in the current area can be indirectly inferred; the feedback values ​​for a smooth sheet are significantly different from those for a thick, soft down comforter. Furthermore, by combining the robot's posture sensor data and path history, it can intelligently identify indentations, persistent wrinkles, or edge areas near the bed. This information is crucial for determining the spray penetration depth, coverage area, and splash prevention strategy.

[0046] In the dynamic spraying decision-making process, the spraying system treats the bed surface as a whole composed of sub-regions of varying quality and applies the most suitable action to each sub-region. A pre-set or learned decision rule library within the controller maps the bound state information to specific spraying action parameters. For example, for areas with high levels of contamination, the decision is to increase the spray dosage by extending the opening pulse width of the solenoid valve or increasing the duty cycle of the micro-pump; for micro-environment areas identified as having high porosity, the decision is to control the piezoelectric atomizing plate to produce finer droplets and may instruct a micro-servo motor to adjust the nozzle to the optimal incident angle; for edge areas, the decision is to close the outer nozzles or narrow the spray angle. This dynamic adjustment, tailored to each location, ensures that the agent achieves optimal adhesion and penetration in bed surface areas with different characteristics.

[0047] To overcome the limitations of traditional open-loop systems, this invention further introduces a pre-action and feedback mechanism. While the dynamic retrospective spraying process is executed, sensors strategically placed behind the spraying components—such as metal oxide semiconductor VOC sensors for detecting volatile components of specific acaricides, or specific wavelength fluorescence sensors used when a safety tracer is added—collect immediate feedback signals of the sprayed effect. This signal reflects the initial concentration or uniformity of the pesticide on the bed surface. The system then performs a high-fidelity correlation between this feedback signal and the coordinates that triggered the spraying, the original state information associated with that point, and the actual spraying action parameters, storing this as an empirical data tuple in non-volatile memory. This step accumulates valuable practical data for the system's self-optimization.

[0048] Based on continuously accumulated feedback data, this invention designs a dynamic optimization process, enabling the system to evolve. The system periodically activates a built-in strategy optimization module, for example, after completing N full-bed applications or when the data volume reaches a certain threshold. This module analyzes stored historical data tuples, with the core objective of optimizing the aforementioned decision mapping relationship. For example, lightweight machine learning algorithms, such as Bayesian optimization algorithms, can be used to continuously try and evaluate, seeking combinations of spraying action parameters that produce better feedback under specific combinations of dirt and microenvironment information, thereby iteratively updating the decision rule base or parameter table. This allows the spraying strategy to continuously improve itself as user experience grows, gradually adapting to the user's specific bedding and lifestyle habits.

[0049] What is particularly innovative is that the optimization mechanism of this invention achieves cross-time-cycle collaboration. The optimized decision mapping relationship will be directly applied to the robot's next spraying task, achieving a gradual improvement in performance. More importantly, the optimization process not only adjusts the spraying itself but also feeds back into the initial cleaning operation. By analyzing historical data, the strategy optimization module can identify a specific type of microenvironment area where spraying effects are consistently poor. It can then generate optimization instructions and transmit them to the path planning and cleaning control module. During the next initial cleaning operation, the robot will adaptively adjust its operating parameters while planning the first planned path or moving in that type of area. For example, it can reduce its movement speed to enhance the tapping and suction effects, increase the roller brush speed, or use a denser sub-bow-shaped path for secondary coverage in that area. The aim is to proactively optimize the microenvironment or dirt information of the area by changing the intensity of the cleaning phase, thereby creating more favorable conditions for subsequent spraying phases. This forms a deep coupling between spraying effect and cleaning strategy optimization, pursuing the maximization of overall mite removal efficiency at the system level.

[0050] For the unique working environment of a bed surface, in step S1, when the robot identifies the bed edge boundary using its surrounding infrared cliff sensors or ultrasonic sensors, the control strategy does not immediately reverse. Instead, it performs a contour cleaning along the bed edge to ensure no boundary is missed. When constructing the information-enhanced path, a special boundary area identifier is assigned to the coordinate points near these bed edges. During the backtracking spraying in step S2, when the robot re-reaches these boundary area identifiers, the spray controller invokes the edge safety spraying subroutine. For example, it controls the nozzles to reduce the spray angle towards the outside of the bed, or directly closes the spray array located on the outside of the robot, using only the inner spray nozzles. This effectively prevents the mite-repellent liquid or droplets from spraying out of the bed, causing pollution and waste, and improves the system's safety and reliability.

[0051] The bed cleaning robot in this embodiment mainly includes the following hardware units: the robot body, which has a main controller (such as an STM32 series or higher-performance ARM processor); the drive unit, which includes two independently driven drive wheels and omnidirectional wheels, and is equipped with an encoder; the cleaning unit, which includes a high-speed roller brush, a beater, and a centrifugal fan forming a dust collection system; the spraying unit, which includes a liquid storage tank, a micro metering pump, a solenoid valve, and a multi-hole atomizing nozzle, the height and angle of which can be adjusted by a micro servo motor; the sensing unit, which includes a gyroscope, an accelerometer, and a forward ultrasonic sensor for mapping and positioning, multiple sets of infrared photocell sensors for cliff detection, as well as a near-field vision sensor, a duct particle sensor, a drive motor current detection circuit, and a VOC sensor located behind the nozzle, all unique to this invention; and a power supply and storage unit.

[0052] The software implementation process of the method of the present invention on the robot is as follows:

[0053] 1. Initialization and Mapping: The user places the robot on the bed surface and starts the device. The robot first performs rapid environmental exploration and mapping. It rotates and probes the bed surface, uses ultrasonic sensors to determine the orientation of the head of the bed, and measures the length and width of the bed surface by moving along the edge combined with encoders and cliff sensors. It constructs a two-dimensional grid map or geometric contour map of the bed surface and plans a bow-shaped path covering the entire bed as the first planned path P1.

[0054] 2. Initial Cleaning and Information Augmentation Path Construction: The robot begins its initial cleaning operation, moving along path P1 at a constant speed while the vacuuming and tapping units work simultaneously. During this process, the information acquisition thread runs concurrently:

[0055] Visual dirt information acquisition: A near-field vision sensor captures images of the bed surface below at a fixed frequency. The image processing module extracts local binary pattern features or grayscale statistical features from each frame and compares them with features from a clean state to generate a visual dirt index V from 0 to 100.i .

[0056] Particulate concentration information acquisition: Real-time readings from the laser particulate sensor inside the air duct are smoothed and filtered, and then recorded as the current particulate concentration value D. i .

[0057] Micro-environment information acquisition: The main controller reads the real-time current value I of the left and right drive motors. left and I right When the robot travels in a straight line on a flat, hard surface, the current value remains stable around the baseline value I0. However, when it enters a fluffy bedding area, the drive wheels sink, increasing resistance and causing a significant rise in the current value. Monitoring the average current (I0)... left +I right The offset ΔI of ) / 2 relative to I0 can be mapped to a fluffiness evaluation value F. i Simultaneously, based on the real-time pose and the preset bed surface map, it can be determined whether the current position is within the preset edge buffer zone, and a boundary region label B can be assigned to that point. flag .

[0058] Information binding: The robot localization module provides the precise pose (x, y, θ) at the current moment. The controller then uses the vector S acquired at this moment... i = (V i D i , F i B flag The information is bound to the pose point and stored in an ordered list L of augmented information paths. Each element of list L is a structure (coordinates (x, y), pose θ, state vector S).

[0059] 3. Dynamic Backtracking Spray Path Generation and Execution: After the initial cleaning, the spraying unit is ready. The system generates a backtracking path P2 based on the path list L. P2 can be an ordered connection of all points in L, or a simplified path that samples L at equal intervals while ensuring coverage. The robot begins to move along P2.

[0060] When the robot localization system determines that it has reached or is approaching the k-th path point P in list L k At that time, the controller reads the state vector S bound to that point. K .

[0061] Decision-making process: An initial decision matrix is ​​preset within the controller. This matrix defines the mapping relationship between state information and spraying parameters, and can be configured as follows:

[0062] If D K Threshold D th If so, the base spraying dose will increase by 50%.

[0063] If F K Threshold F th Then, control the frequency of the atomizing plate to produce fine mist with a diameter of <50μm, and control the servo motor to raise the nozzle by 5mm.

[0064] If B flag If true, then the two outermost nozzles will be closed.

[0065] According to S K By comprehensively querying the decision matrix, the final spraying instruction set is calculated, including the metering pump operating time t. pump Atomization Mode M fog Nozzle posture A, etc.

[0066] The spraying assembly executes this instruction set in P k Complete a customized spraying action near the point.

[0067] 4. Effect Feedback and Data Recording: After the spraying command is executed, the VOC sensor located behind the nozzle will detect a sudden increase in the concentration of volatile components of the pesticide. The controller records this peak concentration E. k This serves as immediate feedback on the spraying effect. Subsequently, the complete data tuple (P) of this operation will be used as feedback. k , S k , (t) pump M fog , A...), E k The data is appended and stored in the historical experience database H.

[0068] 5. Periodic Strategy Optimization: An offline optimization is triggered when the number of data entries in the historical database H exceeds 1000. The optimization algorithm uses a context-based slot machine approach.

[0069] Let the state vector S(V, D, F, B) be... flag Combinations of ) are defined as different contexts.

[0070] Selectable spraying actions (different t) pump M fog The combination is defined as an arm.

[0071] The normalized value of the effect feedback E is used as the reward.

[0072] The goal of the algorithm is to find the arm that yields the highest expected reward for each context.

[0073] After iterative calculations, the algorithm outputs an updated optimal decision matrix that covers all observed contexts. This new matrix will replace the old decision matrix for the next full job run.

[0074] 6. Cross-cycle Co-optimization example: When analyzing the data, the optimization algorithm found that the current strategy was ineffective for all contexts where F_i values ​​were very high and E_k values ​​were consistently low. The optimization module not only tried assigning different spraying actions to these contexts in the new decision matrix—for example, finer atomization combined with a longer action time—but also generated a co-optimization instruction and sent it to the cleaning planning module. This instruction stated: "When F..." i Threshold F thhig When h is reached, it is identified as an "ultra-fluffy challenge zone". During the first cleaning phase of the next operation, when the robot re-enters such an area, the cleaning planning module will temporarily intervene, controlling the robot to reduce its movement speed to 50% of normal and increase the power of the vacuum fan by one level, in order to more thoroughly agitate and clean the deep fibers in the area. This may change the state information S_i collected in the next operation, creating better conditions for the spraying phase.

[0075] Through the above specific implementation methods, the present invention realizes a complete closed loop from environmental perception, information fusion, intelligent decision-making, precise execution to effect evaluation and self-evolution, providing a highly intelligent, adaptive and continuously optimized deep mite removal solution for bed surfaces.

[0076] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for controlling the spraying of a mite remover by a bed surface cleaning robot based on path backtracking, characterized in that, include: Step S1, Perform initial cleaning and construct information-enhanced path: Control the cleaning robot to perform initial cleaning on the bed surface according to the first planned path, and simultaneously collect multimodal state information of the bed surface, and bind the multimodal state information with the coordinate points on the first planned path to form an information-enhanced path; Step S2, Dynamic backtracking spraying based on information-enhanced path: After the first cleaning operation is completed, the cleaning robot is controlled to move along the backtracking path associated with the first planned path. When it moves to the coordinate point on the backtracking path, the multimodal state information corresponding to the coordinate point bound in step S1 is read, and the spraying action of the rear mite spraying component is dynamically adjusted based on the information.

2. The method of claim 1, wherein the method further comprises: The multimodal state information includes at least two types of heterogeneous information extracted from the bed surface environment: The first category is dirt and grime information that characterizes the cleanliness requirements of the bed surface; The second category is microenvironmental information that characterizes the impact of the physical morphology of the spraying bed on spraying operations.

3. The method of claim 2, wherein the method further comprises: The collection of the dirt information includes: Images of the fabric surface are acquired by a near-field vision sensor placed near the cleaning component, and the visual dirt index is obtained through image processing. And / or, the dust concentration value of the inhaled gas is obtained in real time by a particulate sensor integrated in the suction channel.

4. The method of claim 2, wherein the method further comprises: The collection of microenvironment information includes: The fluffiness or stiffness of the bed fabric can be indirectly assessed by pressure sensors or drive motor current feedback. And / or, by using the robot's posture sensors in conjunction with path information, identify depressions, wrinkles, or edge areas on the bed surface.

5. The method of claim 1, wherein: The dynamic adjustment of the spraying action of the post-acaricide spraying component based on this information includes: For coordinate point areas that are bound to different levels of dirt information, the spraying dosage is controlled differently; For coordinate point areas that are bound to different microenvironment information types, the atomization particle size, spray angle, or nozzle height of the spray can be controlled differently.

6. The method of claim 1, wherein: It also includes step S3: In the dynamic retrospective spraying step, a sensor located behind the spraying assembly collects real-time feedback signals reflecting the adhesion of the mite remover to the bed surface. The real-time effect feedback signal is associated and stored with the current coordinate point, its bound multimodal state information, and the spraying action performed.

7. The method of claim 6, wherein the method further comprises: It also includes step S4: Based on the accumulated associated data in step S3, the decision mapping relationship from multimodal state information to spraying actions is periodically optimized and updated using machine learning algorithms, so that the spraying actions tend to be better based on historical effect feedback.

8. The method of claim 7, wherein the method further comprises: Step S4 further includes cross-cycle strategy iteration and cleaning-spraying synergistic optimization, specifically including: The optimized and updated decision mapping relationship is applied to the dynamic backtracking spraying step when the robot executes the method of claim 1 again; And / or, based on specific microenvironment area types identified in historical data where the spraying effect feedback is consistently below a preset threshold, when the robot performs its first cleaning and information binding step again, the robot adaptively adjusts the moving speed, cleaning intensity, or local coverage density of the first planned path for performing the first cleaning operation in such areas.

9. The method of claim 1, wherein: Step S1 also includes: When the edge of the bed is detected by the cliff sensor, the robot is controlled to perform contour cleaning along the edge of the bed, and when constructing the information-enhanced path, special boundary area markers are bound to the coordinate points near the edge of the bed. In step S2, when tracing back to the coordinate point bound by the boundary area marker, the spraying assembly is controlled to reduce the spray angle or close the spray nozzles located on the outside of the bed edge to prevent the acaricide from being sprayed outside the bed.

10. A method for controlling the spraying of a mite-removing agent by a bed surface cleaning robot based on path backtracking, as described in any one of claims 1-9, characterized in that: The first planned path is a bow-shaped path covering the bed surface, and the backtracking path is a guide path that partially or completely overlaps with the bow-shaped path after sampling or simplification.

Images