Toilet cleaning path planning method and system based on visual recognition

Abstract

The application discloses a toilet cleaning path planning method and system based on visual recognition. The method comprises the following steps: acquiring multi-position visual images to extract spatial distribution information of a target area, performing image segmentation processing to form a region-position feature matrix; performing visual recognition based on the feature matrix to obtain region coverage priority classification, performing nonlinear weight distribution according to the area of the region to generate a coverage weight mapping library and determine a key coverage area; identifying a target dense position and an obstacle shielding position, determining the influence weight of a peripheral navigable area based on the obstacle position, and constructing a navigable space weight atlas; acquiring mobile accessibility information based on the weight atlas, identifying a path redundant area, eliminating redundant nodes to generate an optimal navigation path area; forming a robot moving track through the optimal navigation path area, performing visual detection on the covered area to determine a coverage omission area and generate a supplementary coverage path control instruction, and realizing intelligent path planning and adaptive cleaning of the cleaning robot.

Description

Technical Field

[0001] This invention relates to the field of intelligent cleaning robot technology, and in particular to a method and system for planning cleaning paths in restrooms based on visual recognition. Background Technology

[0002] Restroom cleaning in public places and commercial buildings is a crucial aspect of ensuring environmental hygiene. Traditional manual cleaning methods suffer from high labor intensity, low efficiency, and inconsistent quality. While the development of cleaning robot technology has led to the application of automated cleaning equipment in restrooms, existing cleaning robots primarily rely on preset, fixed cleaning patterns, lacking the ability to perceive and adapt to actual environmental conditions. Restroom environments contain various fixed facilities and moving obstacles, and different areas exhibit significant variations in contamination levels. Fixed cleaning paths struggle to adapt to dynamically changing cleaning needs, resulting in low cleaning efficiency and resource waste.

[0003] Existing cleaning path planning methods typically employ a full-coverage strategy, with the robot traversing all areas according to a predetermined trajectory. This approach ignores the differences in cleaning priorities between different areas and fails to optimize the movement path based on obstacle distribution and spatial accessibility. More critically, existing methods lack real-time detection and dynamic supplementation mechanisms for cleaning effectiveness. After cleaning is complete, they cannot identify and address missed areas, impacting overall cleaning quality. Therefore, an intelligent cleaning path planning method is urgently needed to solve at least one of the aforementioned problems. Summary of the Invention

[0004] This invention provides a method and system for planning toilet cleaning paths based on visual recognition. It aims to acquire environmental images in real time and extract spatial distribution features of target areas for image segmentation to form regional feature descriptions. It evaluates cleaning priorities through visual recognition and determines key cleaning targets through nonlinear weight allocation. It constructs a comprehensive weight map by identifying obstacle locations and target distribution features to analyze spatial accessibility. It generates optimized navigation paths through mobile accessibility analysis and redundant node removal. It identifies missed areas and generates supplementary paths through cleaning trajectory monitoring and coverage effect detection. This enables intelligent path planning and adaptive cleaning for cleaning robots.

[0005] The first aspect of this invention proposes a method for planning toilet cleaning paths based on visual recognition, comprising the following steps: Real-time acquisition of multi-location visual image data of the environment; extraction of target region spatial distribution information from the multi-location visual image data; and image segmentation processing of the target region spatial distribution information to form a region-location feature matrix. Based on the region-location feature matrix, visual recognition is performed to obtain the region coverage priority classification. The region coverage priority classification is non-linearly weighted according to the region area to generate a coverage weight mapping library. The coverage weight mapping library is matched and associated with the spatial distribution information of the target region to determine the key coverage area. The key coverage area is used to identify dense target locations and obstacle occlusion locations. The influence weight of the surrounding navigable area is determined based on the obstacle occlusion locations. The influence weight is combined with the dense target locations to construct a navigable spatial weight map. Based on the navigable spatial weight map, mobile accessibility information is obtained, and redundant paths are identified from the mobile accessibility information. Redundant nodes in the redundant paths are removed to generate the optimal navigation path area. The robot's autonomous movement trajectory is formed through the optimal navigation path area. Based on the robot's autonomous movement trajectory, real-time visual detection is performed on the covered area to determine the missed coverage area. The missed coverage area is compared and verified to generate supplementary coverage path control instructions.

[0006] A second aspect of this invention proposes a visual recognition-based restroom cleaning path planning system, comprising: The data acquisition module is used to acquire multi-location visual image data of the environment in real time, extract spatial distribution information of the target area from the multi-location visual image data, and perform image segmentation processing on the spatial distribution information of the target area to form a region-location feature matrix. The region identification module is used to obtain the region coverage priority classification by visual recognition based on the region-location feature matrix, to generate a coverage weight mapping library by non-linear weight allocation of the region coverage priority classification according to the region area, and to match and associate the coverage weight mapping library with the spatial distribution information of the target region to determine the key coverage area. The weight construction module is used to identify the dense locations of targets and the locations of obstacles through the key coverage area, determine the influence weight of the surrounding navigable area based on the location of obstacles, and combine the influence weight with the dense locations of targets to construct a navigable spatial weight map. The path optimization module is used to obtain mobile accessibility information based on the navigable spatial weight map, identify redundant areas in the path from the mobile accessibility information, and remove redundant nodes in the redundant areas to generate the optimal navigation path area. The execution control module is used to form the robot's autonomous movement trajectory through the optimal navigation path area, perform real-time visual detection on the covered area based on the robot's autonomous movement trajectory to determine the covered areas that are missing, and compare and verify the covered areas to generate supplementary coverage path control instructions.

[0007] The beneficial effects of this invention are reflected in the following points: First, by prioritizing regional coverage and grouping by area size to obtain area classification results, differences in cleaning difficulty are identified to determine difficulty classification points. These difficulty classification points are then converted into weight gain factors through threshold segmentation mapping. An exponential mapping method is used to calculate the gain factors and establish a factor ratio table, realizing a nonlinear weight allocation mechanism based on the coupling of area scale effect and cleaning difficulty. A coverage weight mapping library that comprehensively considers pollution level, surface material, and area is established. Second, by dividing obstacle obstruction locations into fixed facility areas and moving object areas, cleaning radius detection signals are transmitted from fixed facility areas to moving object areas, and avoidance distance data at the detection termination position is obtained. A distance parameter table is formed through distance segmentation identification and periodic repetitive attribute extraction. Influence weights are determined based on the area with the largest reachable cleaning range, revealing the obstacle's limiting mechanism on navigation capability. Simultaneously, by analyzing the stain aggregation characteristics of densely populated target locations to construct an aggregation relationship diagram, spatial proximity is measured, and sparse area mining is performed to obtain the synergistic effect of sparse areas. The synergistic effect is integrated with the influence weights to construct a navigable spatial weight map, achieving a comprehensive assessment of cleaning value and accessibility. Finally, by performing repetition detection on redundant areas of the path to obtain round-trip movement records, frequency detection was performed on these records to identify high-frequency points, repetition interval attributes were extracted, and redundancy levels were marked to obtain candidate redundant locations. Movement energy consumption assessment was used to filter and retain nodes while eliminating other redundant nodes, generating the optimal navigation path area with complete coverage and the shortest movement distance. Simultaneously, by performing clean coverage analysis on covered areas to locate under-coverage areas, identifying coverage drop segments and using drop markers to track visual omissions, a set of suspected omission locations was compiled and coverage omission areas were determined. A coverage deviation mapping table was generated by comparing with a standard coverage template to locate path blanks. Path filling processing was then performed to generate a set of filled path segments, which were merged and arranged into supplementary coverage path control commands, achieving automatic identification and supplementary coverage of clean omission areas.

[0008] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0009] The accompanying drawings illustrate specific examples of the technical solutions described in this invention and, together with the detailed embodiments, form part of the specification, serving to explain the technical solutions, principles, and effects of this invention.

[0010] Unless otherwise specified or defined, the same reference numerals in different figures represent the same or similar technical features, and different reference numerals may be used to represent the same or similar technical features.

[0011] Figure 1This is a flowchart illustrating the visual recognition-based restroom cleaning path planning method of the present invention.

[0012] Figure 2 This is a structural block diagram of the visual recognition-based restroom cleaning path planning system of the present invention. Detailed Implementation

[0013] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application can also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0014] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0015] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0016] The technical solutions of the embodiments of this application will be described below.

[0017] like Figure 1 As shown, this embodiment of the invention provides a method for planning a toilet cleaning path based on visual recognition, including the following steps S110-S150: Step S110: Real-time acquisition of multi-location visual image data of the environment; extraction of spatial distribution information of the target area from the multi-location visual image data; and image segmentation processing of the spatial distribution information of the target area to form a region-location feature matrix.

[0018] Specifically, real-time multi-location visual image data of the environment is acquired. Multiple visual acquisition devices are deployed in the restroom environment, including both fixed and mobile cameras. Fixed cameras are installed on the ceiling and in corners, covering key areas such as the sink, toilet area, floor, and doorway. Mobile cameras are mounted on a cleaning robot, acquiring images from different perspectives as the robot moves. Image acquisition parameters are set, including resolution, acquisition frequency, and exposure time. The resolution is set to 1920×1080 pixels to ensure image clarity, the acquisition frequency is set to 3 frames per second to ensure the continuity of dynamic scenes, and the exposure time is automatically adjusted according to ambient lighting. Real-time multi-location visual image data of the environment is acquired, and the restroom status is continuously monitored during the acquisition process. The multi-location visual image data includes image frames captured by different cameras at different times. Each frame includes metadata indicating the acquisition location, shooting angle, acquisition time, and device number. Fixed cameras capture top-down images from the top, while mobile cameras capture side-view images, providing comprehensive coverage of the restroom environment. The acquired multi-location visual image data is transmitted to the processing unit in real time. The transmission uses a wired or wireless network connection, and the transmission delay is controlled within 100 milliseconds to ensure real-time performance.

[0019] Spatial distribution information of target regions is extracted from multi-position visual image data. The received multi-position visual image data undergoes preprocessing, including image denoising, brightness equalization, and distortion correction. Denoising eliminates random noise points in the image, brightness equalization adjusts the brightness consistency of the image under different lighting conditions, and distortion correction corrects image distortion caused by the camera lens. A target detection algorithm is then used to identify target regions in the preprocessed multi-position visual image data. Target regions include surfaces requiring cleaning, such as floors, walls, sinks, mirrors, and sanitary facilities. The target detection algorithm is implemented based on a deep learning model, trained using a labeled dataset of a bathroom scene. The algorithm outputs the bounding box coordinates and region category of the target region, selecting detection results with a confidence level higher than a set threshold as valid target regions. The position and area of ​​the target region in the image coordinate system are extracted; the position is represented by the coordinates of the center point of the bounding box, and the area is calculated from the width and height of the bounding box. The image coordinates are converted to actual spatial coordinates, based on the camera calibration parameters and perspective transformation matrix attached to the multi-position visual image data. The center point coordinates (400, 300) of a certain area in the image correspond to a physical location 2.5 meters away from the camera and 0.8 meters to the left. The extracted spatial distribution information of the target area includes the area category, spatial location coordinates, and area, forming a spatial distribution description of the target area.

[0020] Image segmentation is performed on the spatial distribution information of the target region to form a region-location feature matrix. Each target region is located and further segmented based on its recorded location and boundaries in the spatial distribution information. A semantic segmentation algorithm is used to classify each target region at the pixel level, identifying different surface materials, stain distributions, and areas of cleaning difficulty. The semantic segmentation algorithm assigns a category label to each pixel within the target region, including labels such as "clean surface," "light stains," "heavy stains," and "obstacles." The segmentation results for the washbasin surface show that the central area is a clean surface, the edge areas have light stains, and the area around the faucet has heavy limescale. Based on the segmentation results, the target region is divided into sub-region units, following the principle of consistency between adjacent pixel categories. Consecutive "lightly stained" pixel areas are merged into one sub-region unit, and consecutive "heavyly stained" pixel areas are merged into another sub-region unit. The spatial location, area, and degree of contamination of each sub-region unit are calculated using the spatial coordinates in the target region's spatial distribution information, with the degree of contamination assigned according to the severity of the segmentation category. Construct a region-location feature matrix. The rows of the region-location feature matrix represent the sub-region unit numbers, and the columns represent the feature dimensions. The feature dimensions include spatial location coordinates (x, y), region area S, pollution level P, surface material type M, and cleaning priority L. The element R[i, j] of the region-location feature matrix records the j-th feature value of the i-th sub-region unit, forming a structured description of the region features.

[0021] Step S120: Based on the region-location feature matrix, visual recognition is performed to obtain the region coverage priority classification. The region coverage priority classification is non-linearly weighted according to the region area to generate a coverage weight mapping library. The coverage weight mapping library is matched and associated with the spatial distribution information of the target region to determine the key coverage area.

[0022] Specifically, priority grading of area coverage is obtained through visual recognition based on the region-location feature matrix. Feature data for each sub-region unit is extracted from the region-location feature matrix, including the degree of contamination, surface material type, and area. Visual recognition analysis is performed on the feature data in the region-location feature matrix, and the recognition process assesses the cleaning demand intensity of each sub-region. The cleaning demand intensity is jointly determined by the degree of contamination and the surface material; areas with higher contamination levels have higher demand intensity, and surfaces prone to dirt accumulation also have higher demand intensity. The cleaning demand intensity index is calculated based on the region-location feature matrix using the formula I_need=P×M_factor, where I_need is the cleaning demand intensity, P is the degree of contamination (0-100), and M_factor is the material influence factor (1.0 for tile surfaces, 1.2 for stainless steel surfaces, and 0.8 for glass surfaces). The stainless steel faucet area near the washbasin has a contamination level of 75, a material factor of 1.2, and a cleaning demand intensity of 90. Sub-regions are prioritized according to their cleaning demand intensity, with higher intensity values ​​indicating higher priority. Sub-regions are divided into three levels: high priority, medium priority, and low priority. Sub-areas with a cleaning demand intensity greater than 70 are classified as high priority, those with an intensity between 40 and 70 as medium priority, and those with an intensity less than 40 as low priority. A regional coverage priority classification is obtained, and the priority level of each sub-area is recorded.

[0023] In some embodiments, the step of generating a coverage weight mapping library by nonlinearly weighting the regional coverage priority classification according to the regional area includes: grouping the regional coverage priority classification according to the area size to obtain area classification results; identifying cleaning difficulty differences and determining difficulty classification points based on the area classification results; performing threshold segmentation mapping on the difficulty classification points to convert them into weight gain factors to form a factor ratio table; and establishing a coverage weight mapping library based on the factor ratio table.

[0024] The area coverage priority classification is grouped by size to obtain area tiers. Area data for each sub-area is extracted from the area coverage priority classification, with the data in square meters. Statistical analysis is performed on the area data from the area coverage priority classification to calculate the minimum, maximum, average, and distribution range of the area. In the restroom environment, the minimum sub-area area is 0.1 square meters, the maximum is 6 square meters, and the average is 1.2 square meters. Grouping intervals are set based on the area distribution characteristics, dividing the area range into several tiers. Tiering uses an unequal interval method, with smaller area intervals being denser and larger area intervals being more sparse, reflecting the need for refined management of small areas. The area tier standards are set as follows: micro-area 0-0.5 square meters, small area 0.5-1.5 square meters, medium area 1.5-3 square meters, large area 3-5 square meters, and extra-large area 5 square meters or more. Each sub-area in the area coverage priority classification is assigned to the corresponding tier interval according to its area size, comparing the sub-area area with the tier interval boundaries during the allocation process. The faucet area of ​​the washbasin, with an area of ​​0.3 square meters, is assigned to the "small area" category, and the floor area, with an area of ​​2.5 square meters, is assigned to the "medium area" category. The area grading results are obtained, and the results record the area grading level and level number of each sub-area, forming a sub-area classification grouped by area.

[0025] The cleaning difficulty differences were identified and difficulty grading points were determined based on the area grading results. The cleaning difficulty characteristics of different area grading levels were analyzed based on the area grading results. Cleaning difficulty is affected by area size, shape complexity, and accessibility. Larger areas require longer cleaning equipment travel distances, areas with complex shapes require frequent adjustments to the cleaning posture, and areas with poor accessibility require additional operational preparation. The cleaning time and efficiency of sub-areas within each area grading level were statistically analyzed. Cleaning time was obtained from historical cleaning records or experimental measurements. Small areas had an average cleaning time of 30 seconds and high efficiency; medium areas had an average cleaning time of 60 seconds and moderate efficiency; medium areas had an average cleaning time of 120 seconds and efficiency began to decline; and large areas had an average cleaning time of 200 seconds and significantly reduced efficiency. Differences in cleaning difficulty were identified based on the area grading results, which manifested as a stepwise decrease in cleaning efficiency. The cleaning efficiency ratio between adjacent area grading levels was calculated; a ratio less than 0.8 indicated a significant increase in difficulty. The efficiency ratio between medium and small areas was approximately 0.75, and the efficiency ratio between large and medium areas was approximately 0.70, both showing a significant decrease. Difficulty grading points were determined, corresponding to the area boundary values ​​where cleaning efficiency significantly decreased. The identified difficulty grading points were 0.5 square meters, 1.5 square meters, and 3 square meters, marking the points where cleaning difficulty increased dramatically.

[0026] The difficulty grading points are mapped to weighted gain factors using threshold segmentation to form a factor allocation table. Threshold segmentation is performed based on the identified difficulty grading points. Threshold segmentation divides the area range into multiple segments according to the difficulty grading points, with each segment corresponding to a difficulty level. Segment 1 (0-0.5 square meters) corresponds to difficulty level 1, segment 2 (0.5-1.5 square meters) corresponds to difficulty level 2, segment 3 (1.5-3 square meters) corresponds to difficulty level 3, and segment 4 (over 3 square meters) corresponds to difficulty level 4. Each difficulty level is mapped to a weighted gain factor, which reflects the relative gain in cleaning workload at that difficulty level. The higher the difficulty level, the larger the gain factor, indicating that higher-difficulty areas require more cleaning resources. The gain factor is calculated using an exponential mapping method based on the segments divided by the difficulty grading points, with the formula G_factor = 1.0 × 1.5^(L-1), where G_factor is the weighted gain factor, and L represents the difficulty level value (1-4). The gain factor for difficulty level 1 is 1.0, for difficulty level 2 it is 1.5, for difficulty level 3 it is 2.25, and for difficulty level 4 it is 3.375. A factor allocation table is generated, organized in tabular form, listing area segments, difficulty levels, and corresponding weighted gain factors. The factor allocation table supports quick lookup of gain factors based on sub-region area.

[0027] A coverage weight mapping library is established based on the factor allocation table. The records of each sub-region in the regional coverage priority classification are traversed, and the corresponding weight gain factor is retrieved from the factor allocation table based on the sub-region area. Based on the weight gain factor provided by the factor allocation table, the comprehensive coverage weight is calculated by combining the sub-region's priority base weight. The comprehensive weight is the product of the priority base weight and the weight gain factor, calculated using the formula W_total = W_priority × G_factor, where W_total is the comprehensive coverage weight, W_priority is the priority base weight (3 for high priority, 2 for medium priority, and 1 for low priority), and G_factor is the weight gain factor. For a high-priority, medium-priority region with an area of ​​2 square meters, the priority base weight is 3, the gain factor is 2.25, and the comprehensive coverage weight is 6.75. A coverage weight mapping library is established, stored using a database or hash table structure. The primary key is the sub-region number, and the fields include the sub-region number, priority level, region area, weight gain factor, and comprehensive coverage weight. The coverage weight mapping library integrates priority and area weight information to form complete coverage weight data.

[0028] The key coverage areas are determined by matching and associating the coverage weight mapping library with the spatial distribution information of the target area. The process involves acquiring the spatial distribution information of the target area and the coverage weight mapping library data. The spatial distribution information of the target area includes the spatial coordinates and boundaries of each sub-area, while the coverage weight mapping library contains the comprehensive coverage weight of each sub-area. Matching and association processing is performed, with the association based on the sub-area number to establish a one-to-one correspondence between the spatial distribution information of the target area and the coverage weight mapping library. Each sub-area in the spatial distribution information of the target area is associated with the corresponding record in the coverage weight mapping library, adding a comprehensive coverage weight attribute to the sub-area. The associated data includes the spatial location, boundaries, contamination level, priority level, and comprehensive coverage weight of the sub-area, forming a complete description of the area's characteristics. Key coverage areas are selected based on the comprehensive coverage weight, with a weight threshold set as the selection criterion. Sub-areas with a comprehensive coverage weight greater than the threshold are determined to be key coverage areas. The threshold is dynamically adjusted based on cleaning resources and time constraints. For example, the comprehensive coverage weight of heavily soiled areas around sink faucets is 9.0, the weight of soiled areas inside toilets is 8.5, and the weight of waterlogged areas on the floor is 7.2, all exceeding the threshold of 6.0 and thus being determined as key coverage areas. A set of key coverage areas is determined, which includes all sub-areas that need to be prioritized for cleaning, along with their spatial location and weight information, providing a basis for target areas in cleaning route planning.

[0029] Step S130: Identify dense target locations and obstacle occlusion locations through key coverage areas, determine the influence weight of the surrounding navigable area based on the obstacle occlusion location, and combine the influence weight with the dense target locations to construct a navigable spatial weight map.

[0030] Specifically, key coverage areas are used to identify densely populated target locations and obstacle-occupied locations. Spatial distribution data of key coverage areas is acquired, including the location coordinates, boundaries, and overall coverage weight of each key coverage sub-region. The spatial clustering characteristics of key coverage areas are analyzed to identify locations where multiple sub-regions are spatially close or overlapping. The Euclidean distance between the center points of sub-regions within key coverage areas is calculated; sub-regions with distances less than a threshold are considered spatially adjacent. The center point distance between three high-priority sub-regions around a washbasin is less than 0.5 meters, forming a spatial cluster. The number of key coverage sub-regions within a certain radius around each spatial location is counted; locations with a higher number are marked as densely populated target locations. Densely populated target locations represent areas with concentrated cleaning tasks, requiring frequent access and operation of cleaning equipment, and are key coverage targets for cleaning path planning. The spatial location information of key coverage areas is used to locate the area where obstacles need to be detected, and obstacles within this area are identified. Obstacles include fixed facilities and moving objects. Fixed facilities include toilets, washbasin bases, corners, pipes, and fixed partitions; moving objects include trash cans, cleaning tools, mop buckets, and temporarily placed items. An object detection algorithm is used to identify the type and location of obstacles in the image, and the bounding box coordinates and category labels of the obstacles are output. The occlusion impact on the cleaning path is determined; occlusion locations are areas that the cleaning equipment cannot directly pass through or reach. The obstacle occlusion location records the spatial coordinates of the obstacle, the occlusion range, the obstacle type, and the occlusion intensity level.

[0031] In some embodiments, determining the influence weight of the surrounding navigable area based on the obstacle occlusion location includes: dividing the obstacle occlusion location into a fixed facility area and a moving object area; transmitting a cleaning radius detection signal from the fixed facility area to the moving object area; acquiring avoidance distance data of the cleaning radius detection signal to form a distance parameter table; and selecting the area with the largest cleaning reach as the influence weight of the surrounding navigable area according to the distance parameter table.

[0032] Obstacle occlusion locations are categorized into fixed facility areas and moving object areas. Obstacle category labels and location information are extracted from these locations. Obstacle occlusion locations are classified based on their fixedness and mobility. Fixed facilities are those installed inside the restroom and whose position remains unchanged, including toilets, sinks, mirrors, pipes, and wall structures. Moving objects are items that may change position or be temporarily placed, including trash cans, mops, tissue boxes, and cleaning supplies. The fixedness of obstacles within the occluded locations and the stability of their position in historical images are determined. A toilet whose position did not change in images over the past week is classified as a fixed facility. A trash can whose position changed in images at different times, with a change range exceeding 0.2 meters, is classified as a moving object. Obstacle occlusion locations classified as fixed facilities are assigned to the fixed facility area, which records the spatial coordinates, dimensions, occlusion range, and facility type of the fixed obstacle. Obstacle occlusion locations classified as moving objects are assigned to the moving object area, which records the current position, movement frequency, possible range of position changes, and object type of the moving obstacle. The distinction between fixed facility areas and moving object areas provides different processing strategies for navigation planning. The obstructions in fixed facility areas are permanent navigation obstacles that need to be completely avoided by the path, while the obstructions in moving object areas are temporary navigation obstacles that can be dynamically adjusted according to the current position of the object.

[0033] Cleaning radius detection signals are transmitted from the fixed facility area to the moving object area. Detection signal transmitters are set at the boundaries of the fixed facility area, evenly distributed along the contours of the fixed facilities within the area. Cleaning radius detection signals are emitted outward from these transmitters, simulating the process of cleaning equipment moving outward from the edge of the fixed facility area. The cleaning radius detection signal carries distance information, recording the cumulative distance from the transmitter to the current position. The detection signal propagates radially, detecting whether it encounters the boundary of an obstacle in the moving object area during propagation. The edge of the sink base serves as the transmitter point in the fixed facility area, emitting a cleaning radius detection signal outward. The signal travels 0.3 meters before encountering a trash can in the moving object area. When the cleaning radius detection signal reaches the boundary of an obstacle in the moving object area, the detection terminates, and the termination position and cumulative distance are recorded. The termination position of the cleaning radius detection signal marks the boundary of the cleanable area between the fixed facility area and the moving object area. Cleaning radius detection signals are emitted from multiple transmitters in the fixed facility area, forming a detection signal network covering the surrounding space. Each cleaning radius detection signal path records the transmitter coordinates, propagation direction, termination position coordinates, and cumulative distance, forming a detection path dataset, providing raw data for avoidance distance calculation.

[0034] For example, the step of obtaining the avoidance distance data of the cleaning radius detection signal to form a distance parameter table includes: identifying the detection termination position from the cleaning radius detection signal; performing distance segmentation identification on the detection termination position to generate segmented distance groups; extracting periodic repetition attributes from the segmented distance groups to obtain a repetition pattern table; and combining the repetition pattern table with the detection termination position to perform distance classification marking to generate a distance parameter table.

[0035] Identify the detection termination position from the cleaning radius detection signals. Acquire the propagation path data of the cleaning radius detection signals, recording the emission point, propagation direction, propagation distance, and termination status. The cleaning radius detection signal terminates when it encounters an obstacle boundary during propagation, and the termination status is marked as "obstacle encounter termination." The cleaning radius detection signal terminates when it propagates to a preset maximum detection distance without encountering an obstacle, and the termination status is marked as "distance termination." Filter the detection signals with the termination status of "obstacle encounter termination" from the cleaning radius detection signal path data. The termination positions of these signals correspond to actual obstacle boundaries, representing the reachable boundaries of the cleaning equipment. Extract the spatial coordinates of the detection termination position, represented in a two-dimensional plane coordinate system. The cleaning radius detection signal emitted from the edge of the sink terminates at coordinates (1.5, 2.3) when it encounters the boundary of the trash can; this coordinate is recorded as the detection termination position. Statistically analyze the detection termination positions of all cleaning radius detection signals to form a set of termination position coordinates, which contains all valid obstacle boundary position information.

[0036] The detection termination position is segmented and identified to generate segmented distance groups. The straight-line distance from each detection termination position to the corresponding launch point is calculated; this straight-line distance is the avoidance distance. The avoidance distance is calculated using the Euclidean distance formula, which is as follows: Where D is the avoidance distance in meters, (x1, y1) are the coordinates of the launch point, and (x2, y2) are the coordinates of the detection termination position. The avoidance distance from the launch point (1.0, 2.0) to the detection termination position (1.5, 2.3) is approximately 0.58 meters. All avoidance distance data were statistically analyzed, including the distribution range, mean, median, and clustering characteristics. The minimum avoidance distance was 0.2 meters, and the maximum was 0.9 meters, with most distances concentrated in the 0.3-0.6 meter range. The median was approximately 0.45 meters. The avoidance distances were segmented based on the clustering characteristics of the distance distribution and the size of the cleaning equipment. The segmentation standards were set as follows: near distance 0-0.3 meters, medium distance 0.3-0.6 meters, far distance 0.6-0.9 meters, and ultra-far distance over 0.9 meters. The avoidance distance for each detection termination position was assigned to a corresponding segment, forming segmented distance groups. The near-range segment contains 18 detection termination positions, the mid-range segment contains 35 detection termination positions, the far-range segment contains 12 detection termination positions, and the ultra-far-range segment contains 3 detection termination positions. Segmented distance groups discretize continuous distance data into a finite number of distance levels, facilitating subsequent pattern extraction and hierarchical processing.

[0037] The periodic repetition attributes of segmented distance groups are extracted to obtain a repetition pattern table. The spatial distribution pattern of the segmented distance groups is analyzed to identify the spatial arrangement of detection termination positions within the same segment. Some obstacle configurations exhibit periodic characteristics, leading to periodic repetition of avoidance distances. In a restroom, the layout of toilets and sinks in multiple stalls is similar, with similar avoidance distances at corresponding positions within each stall, reflecting the periodicity of the spatial layout. The spatial distribution period of detection termination positions in each segmented distance group is statistically analyzed. Detection termination positions in the mid-distance segment cluster every 1.2 meters along the wall, reflecting the periodic layout of the toilet stalls, with a period length of 1.2 meters. The period length and repetition frequency of the spatial distribution are calculated; the period length is the distance between adjacent repetition positions, and the repetition frequency is the frequency of the period in space. Periodic repetition attributes of the segmented distance groups are extracted, including repetition period, repetition direction, repetition frequency, and period stability. A repetition pattern table is obtained, recording the periodic characteristics of each segmented distance group and listing the segment identifier, period length, repetition direction, and repetition frequency. The repeating pattern of the mid-distance segment is 1.2 meters in period, parallel to the wall in direction, repeated 3 times, and has high stability.

[0038] A distance parameter table is generated by combining the repetition pattern table with the detection termination position data to perform distance classification. The repetition pattern table and detection termination position data are integrated. Based on the periodic characteristics in the repetition pattern table, the detection termination positions are grouped. Detection termination positions belonging to the same repetition period are grouped together, representing similar obstacle configurations and avoidance distance patterns. Statistical analysis is performed on the avoidance distance of each group of detection termination positions, calculating the average avoidance distance, distance variance, and distance stability index within the group. A group containing 5 detection termination positions has an average avoidance distance of 0.42 meters and a distance variance of 0.008 square meters, indicating high distance stability and small differences in avoidance distances among positions within the group. Distance is classified based on the average avoidance distance and distance stability. An average distance less than 0.3 meters and high stability is marked as Grade A, indicating a narrow passage; an average distance of 0.3-0.6 meters and high stability is marked as Grade B, indicating a moderate passage; an average distance greater than 0.6 meters is marked as Grade C, indicating a wide passage; and low stability is marked as Grade D, indicating an uncertain area requiring careful handling. Generate a distance parameter table, which integrates detection termination position, avoidance distance, segmentation identifier, repetition pattern, and distance classification marker, organized in a structured table format, supporting fast query and path planning applications.

[0039] Based on the distance parameter table, the area with the largest reachable cleaning range is selected as the influence weight of the surrounding navigable area. Avoidance distance data and distance classification labels for each spatial orientation are extracted from the distance parameter table. The reachable cleaning range for each orientation is calculated as the avoidance distance in the distance parameter table minus the radius and safety margin of the cleaning equipment itself. For an orientation with a cleaning equipment radius of 0.2 meters, a safety margin of 0.1 meters, and an avoidance distance of 0.45 meters in the distance parameter table, the reachable cleaning range is 0.15 meters. The reachable cleaning ranges for different orientations are compared to identify the orientations and areas with the largest reachable ranges in the distance parameter table. The reachable cleaning range in the open area in front of the sink is 0.5 meters, and the reachable cleaning range in the narrow passage to the left of the sink is 0.1 meters; the area in front has the largest reachable range. The area with the largest reachable cleaning range is selected as the preferred direction for the surrounding navigable area, as these directions have the highest navigation efficiency and the strongest passage capability for the cleaning equipment. The impact weight of the surrounding navigable area is calculated. This impact weight is positively correlated with the clean reachability, and the formula is W_impact = R_reach / R_max, where W_impact is the impact weight (0-1), R_reach is the clean reachability in meters, and R_max is the ideal maximum reachability (1.0 meter). An area with a clean reachability of 0.5 meters has an impact weight of 0.5, and an area with a reachability of 0.15 meters has an impact weight of 0.15. The impact weight quantifies the degree to which obstacle occlusion restricts navigation capabilities; a higher weight indicates that the area is more suitable for a clean path, while a lower weight indicates that access is restricted and requires avoidance or careful handling.

[0040] In some embodiments, the step of combining the influence weights with the target dense locations to construct a navigable spatial weight map includes: analyzing the stain aggregation characteristics based on the target dense locations to construct an aggregation relationship map; determining the spatial proximity of the aggregation relationship map to obtain a proximity data table; performing sparse region mining on the proximity data table to obtain sparse region synergy effects; and using the sparse region synergy effects and the influence weights to fuse them into a navigable spatial weight map.

[0041] Based on the analysis of stain aggregation characteristics in densely populated target locations, an aggregation relationship map was constructed. Stain distribution information for each sub-region within these locations was extracted, including stain type, degree of contamination, stain area, and stain density. Spatial aggregation patterns of stains in these locations were analyzed to identify densely and sparsely distributed areas. Stain similarity between sub-regions within these locations was calculated, based on the consistency of stain type and the proximity of contamination degree. The three densely populated target location sub-regions around the sink all exhibited limescale stains with a severe degree of contamination, sharing the same stain type and similar degree, resulting in a similarity of 0.9. The floor sub-region contained dust and footprints, different from the limescale stains on the sink, resulting in a low similarity of only 0.3. Stain aggregation characteristics were identified, including aggregation center, aggregation range, and aggregation intensity. The aggregation center represents the location of the most concentrated stains, the aggregation range is the spatial area of ​​continuous stain distribution, and the aggregation intensity reflects the density of stains within the aggregation area. Construct a clustering graph, which is represented by a graph structure. Nodes represent sub-regions within densely packed target locations, edges represent stain similarity relationships between sub-regions, and edge weights are the stain similarity scores. Closely connected groups of nodes in the clustering graph represent cleaning areas with similar stain characteristics, requiring the same cleaning strategy.

[0042] Spatial proximity data was obtained by determining the spatial proximity of the clustering graph. The spatial coordinates of the nodes were extracted from the clustering graph, derived from the spatial distribution data of densely packed target locations. The spatial distance between nodes in the clustering graph was calculated using Euclidean distance. The distance between the center points of the washbasin sub-region and the mirror sub-region was 0.8 meters, the distance between the washbasin sub-region and the floor sub-region was 1.5 meters, and the distance between the toilet sub-region and the washbasin sub-region was 2.0 meters. The spatial proximity between nodes in the clustering graph was determined. Proximity reflects the closeness of nodes in space; the closer the distance, the higher the proximity, and the farther the distance, the lower the proximity. The proximity was calculated using a Gaussian function, with the formula N_proximity = exp(-d² / 2σ²), where N_proximity is the spatial proximity value (0-1), d is the spatial distance in meters, and σ is the proximity decay parameter (1.0 meter). The proximity of node pairs in the clustering graph is approximately 0.73 at a distance of 0.8 meters, approximately 0.32 at a distance of 1.5 meters, and approximately 0.14 at a distance of 2.0 meters. Spatial proximity is calculated for all node pairs in the clustering graph, forming a proximity matrix of size N×N, where N is the number of nodes. A proximity data table is obtained, which records node pair identifiers, spatial distances, and spatial proximity values ​​in tabular form. This table supports quick querying of the proximity relationships of any node pair, providing fundamental data for sparse area mining.

[0043] This study utilizes sparse region mining to uncover the synergistic effect of sparse regions in a proximity data table. Analyzing the proximity distribution characteristics in the table identifies node pairs with low proximity values. Proximity values ​​below a threshold of 0.4 indicate that nodes are spatially distant, and the areas between these nodes constitute sparse regions. Sparse regions are blank areas between densely populated target locations; stains are sparsely distributed, but cleaning paths are still needed to connect different densely populated target locations. The spatial location and area of ​​sparse regions in the proximity data table are statistically analyzed. Sparse regions are typically located in connecting passages between two densely populated target locations. For example, the floor passage between the sink and toilet densely populated areas is a sparse region, with an area of ​​approximately 1.5 square meters and a proximity of 0.28. The synergistic effect of sparse regions is explored; this effect refers to the ability of a cleaning path to simultaneously serve multiple densely populated target locations when passing through a sparse region. Paths through sparse regions allow for efficient switching between different densely populated areas, improving overall cleaning efficiency and avoiding time wasted by repeated back-and-forth movements. The synergy index for sparse regions is calculated based on the number of densely connected target locations and the length of the connection path within the sparse region. The formula is E_synergy = N_connected / L_path, where E_synergy is the synergy index in terms of connections per meter, N_connected is the number of densely connected locations in the sparse region, and L_path is the path length through the sparse region in meters. For example, a sparse region connecting 3 densely connected locations with a path length of 2 meters has a synergy index of 1.5 connections per meter. A higher value indicates more densely connected locations per unit path length and higher synergy value. Sparse region synergy data is obtained, recording the location, connection relationships, synergy index, and normalized synergy value of the sparse region. Normalization ensures the synergy index ranges from 0 to 1 for easy subsequent fusion calculations.

[0044] A sparse region synergy effect and influence weight are fused to create a navigable spatial weight map. This involves integrating sparse region synergy effect data with influence weight data from surrounding navigable areas. The sparse region synergy effect data reflects the synergistic value of a path passing through a sparse region, while the influence weight data reflects the limitations imposed by obstacles on navigation capabilities. A unified spatial grid system is established, mapping the sparse region synergy effect data and influence weight data to the same grid cells. For each grid cell, its sparse region status is queried, and a normalized synergy effect index is extracted, along with its influence weight value. The synergy effect index for grid cells outside sparse regions is set to a baseline value of 1.0. The fusion weight of a grid cell is calculated, taking into account both the synergistic effect and the influence weight in the sparse region. The formula is W_fused = E_synergy_norm × W_impact × K_balance, where W_fused is the fusion weight (0-1), E_synergy_norm is the normalized synergistic effect index (0-1), W_impact is the influence weight (0-1), and K_balance is a balance coefficient of 0.8 used to adjust the relative importance of the two types of weights. For a certain grid cell in the sparse region, the normalized synergistic effect is 1.0, the influence weight is 0.6, and the fusion weight is approximately 0.48. As a navigable spatial weight map, the map represents the comprehensive navigation value of each location in space in the form of a two-dimensional matrix. The matrix elements are the fusion weights of the grid cells. Locations with high fusion weights have both better accessibility and can efficiently connect multiple cleaning targets, making them preferred areas for cleaning path planning.

[0045] Step S140: Obtain mobile accessibility information based on the navigable spatial weight map, identify path redundancy areas from the mobile accessibility information, and remove redundant nodes from the path redundancy areas to generate the optimal navigation path area.

[0046] Specifically, mobility reachability information is obtained based on a navigable spatial weight graph. The fusion weight data of each grid cell is extracted from the navigable spatial weight graph; the fusion weight reflects the navigation value and passage difficulty of the grid cell. The connectivity between grid cells in the navigable spatial weight graph is analyzed to determine whether the cleaning robot can move from one grid cell to an adjacent grid cell. Connectivity determination is based on whether there are obstacles obstructing the movement between grid cells in the navigable spatial weight graph and whether the passage width meets the robot's passage requirements. Two adjacent grid cells are considered connected when both have a fusion weight greater than a threshold of 0.3 and there are no obstacles between them; otherwise, they are considered disconnected. An reachability network is constructed, represented by a graph structure, where nodes are navigable grid cells, edges represent the connectivity between adjacent cells, and the edge weight represents the movement cost. The movement cost comprehensively considers the fusion weights of the two cells and the movement distance, and is calculated using the formula C_move=d / (W_fused1+W_fused2)×2, where C_move is the movement cost in meters, d is the distance between the center points of adjacent cells in meters, and W_fused1 and W_fused2 are the fusion weights of the two cells, ranging from 0 to 1. With an adjacent cell distance of 0.5 meters and fusion weights of 0.6 and 0.8 respectively, the movement cost is approximately 0.71. Mobility reachability information is obtained, which includes the reachability network structure of the grid cells, node connectivity, and edge movement costs, describing the robot's mobility and efficiency in the restroom environment.

[0047] Redundant areas in the path are identified from mobile accessibility information. A preliminary navigation path is constructed based on this information, generated using a graph search algorithm. The algorithm starts from the starting position, visits all key coverage areas, and returns to the starting position. The preliminary navigation path includes the sequence of traversed grid cells, the direction of movement, and the estimated travel time. The spatial trajectory characteristics of the preliminary navigation path are analyzed to identify areas where the robot repeatedly passes through or detours. Repeated passage refers to the robot passing through the same location multiple times at different times; detours refer to unnecessary twists and turns in the path. These phenomena reduce the utilization efficiency of the accessibility network in the mobile accessibility information. The number of times each grid cell is traversed in the path is calculated; cells traversed more than once are considered repeatedly traversed. For example, a cell around the sink is traversed 3 times, a cell in the center of the ground is traversed 2 times, and a cell at the doorway corner is traversed 1 time. Back-and-forth movement segments are identified in the path. Back-and-forth movement segments refer to situations where the robot moves from position A to position B and then returns to the vicinity of position A, forming an unnecessary repeated trajectory. For example, the robot moves from the sink to the doorway area and then returns to the sink to continue cleaning, forming a back-and-forth movement segment with a distance of approximately 3 meters. Regions that repeatedly traverse the same cell set and regions with frequent round-trip movements are marked as path redundancy regions. Path redundancy regions represent parts of the path planning where there is an efficiency loss; these regions contain redundant movements that can be optimized and are the focus of path optimization.

[0048] In some embodiments, the step of removing redundant nodes from the path redundancy area to generate the optimal navigation path area includes: performing repetition detection on the path redundancy area to obtain round-trip movement records; identifying repeatedly visited points from the round-trip movement records as redundant location candidates; performing mobile energy consumption assessment on the redundant location candidates to filter and retain nodes; and removing the remaining redundant nodes based on the retained nodes to form the optimal navigation path area.

[0049] Repeatability detection is performed on redundant areas to obtain round-trip movement records. Complete trajectory data of the initial navigation path in the redundant areas is extracted, recording the sequence of locations the robot traverses in chronological order and the arrival time of each location. Repeatability detection is performed on the trajectory data in the redundant areas, identifying repeatedly occurring locations. A position coordinate matching method is used to determine if locations are repeated; two locations are considered the same if the coordinate distance is less than a threshold of 0.1 meters. The frequency and timing of each location in the redundant areas are counted; locations with more than one occurrence are considered repeated. The location (2.5, 3.0) in front of the sink appears three times in the redundant area trajectory at minutes 5, 12, and 18, with a repeatability of 3. The movement path between adjacent repeated occurrences is analyzed to extract the robot's movement trajectory from the first occurrence to the second. After the robot first passes in front of the sink, it moves to the ground area, then to the doorway area, and then returns to the sink, forming one round-trip movement. Obtain round-trip movement records, which include the starting and ending positions, the round-trip path trajectory, the round-trip movement distance, and the round-trip movement time, describing the repetitive movement patterns in the redundant areas of the path.

[0050] For example, identifying repeatedly passed points as redundant location candidates from the round-trip movement records includes: performing frequency detection on the round-trip movement records to obtain frequency statistics results; identifying high-frequency points from the frequency statistics results to generate a high-frequency point set; extracting the repetition interval attribute from the high-frequency point set to form an interval attribute list; and marking the redundancy degree according to the interval attribute list to obtain redundant location candidates.

[0051] Frequency statistics are obtained by performing pass-through frequency detection on the round-trip movement records. All relevant location point data, including location coordinates and pass-through time, are extracted from the records. Location points in the records are grouped and clustered according to their spatial coordinates; locations with a coordinate distance of less than 0.1 meters are grouped together, representing the same physical location. The frequency of each location group in the round-trip movement records is calculated; the frequency is the total number of times that location is traversed in the trajectory. The location group in front of the sink contains 3 locations with similar coordinates, with a total frequency of 3. The location group in the center of the ground contains 2 locations, with a total frequency of 2. The location group in the doorway area contains 1 location, with a frequency of 1. The statistical distribution of pass-through frequencies is calculated, including the maximum, minimum, average, and standard deviation, to assess the concentration of location point distribution in the round-trip movement records. In the restroom environment, the maximum pass-through frequency for each location group is 4, the minimum is 1, the average is 1.8, and the standard deviation is 0.6. The frequency statistics are obtained and recorded in tabular form, showing the coordinates, frequency, and ranking of each location group, supporting rapid identification of frequently traversed locations.

[0052] High-frequency points are identified from frequency statistics results to generate a high-frequency point set. The frequency distribution characteristics in the frequency statistics results are analyzed, and a high-frequency threshold is set as the criterion for determining high-frequency points. The high-frequency threshold is determined based on the statistical characteristics of the frequency distribution in the frequency statistics results, typically set as the average frequency plus one standard deviation. With an average frequency of 1.8 and a standard deviation of 0.6, the high-frequency threshold is set to 2.4. Location groups with frequencies greater than or equal to the high-frequency threshold are selected from the frequency statistics results; these location groups are traversed with significantly higher frequencies than the average. The frequency 3 of the location group in front of the handwashing station exceeds the threshold of 2.4, the frequency 3 of the location group in the passageway exceeds the threshold, and the frequency 2 of the location group in the center of the ground is below the threshold. The selected high-frequency location groups are marked as high-frequency points. High-frequency points represent frequently traversed key locations in the path and are the main source of path redundancy. A high-frequency point set is generated, containing the coordinates, traversal frequency, and functional role of all high-frequency location groups in the path. The locations in front of the handwashing station and the location in the passageway are identified as high-frequency points; the high-frequency point set records their coordinates as (2.5, 3.0) and (3.2, 2.5), respectively, with a frequency of 3 for both.

[0053] The repetition interval attribute is extracted from the high-frequency point set to form an interval attribute list. For each high-frequency point in the set, all its elapsed times on the trajectory are extracted. High-frequency points in the set are traversed multiple times on the trajectory, with each elapsed time corresponding to a time marker. The elapsed times of the high-frequency point in front of the handwashing station are at the 5th, 12th, and 18th minutes. The time interval between adjacent elapsed times of each high-frequency point in the set is calculated; the time interval reflects how often the robot repeatedly passes through that location. The first interval for the high-frequency point in front of the handwashing station is 7 minutes, and the second interval is 6 minutes. The distribution characteristics of the time intervals are analyzed to extract the repetition interval attribute. The repetition interval attribute includes the average interval, interval variance, and interval periodicity. The average interval is the arithmetic mean of all intervals; the average interval for the high-frequency point in front of the handwashing station is 6.5 minutes. The interval variance reflects the stability of the intervals; a small variance indicates regular repetition, while a large variance indicates irregular repetition. Periodicity is identified by analyzing the autocorrelation of the interval sequence; strong periodicity indicates that the high-frequency point is repeatedly visited at a fixed time period. A list of interval attributes is generated, which records the average interval, interval variance, and periodicity index of each high-frequency point in the high-frequency point set, describing the repeated access pattern of the high-frequency points.

[0054] Redundancy candidates are obtained by marking redundancy levels according to the interval attribute list. Repeated interval attribute data for each high-frequency point is extracted from the interval attribute list. The relationship between repeated intervals and path redundancy in the interval attribute list is analyzed. High-frequency points with short average intervals and strong periodicity have high redundancy. Short intervals indicate frequent robot trips to the location, and strong periodicity indicates a fixed and repetitive trip pattern; these characteristics reflect the redundancy of the location. The high-frequency point in front of the handwashing station has an average interval of 6.5 minutes and a periodicity index of 0.85, indicating that this location is regularly and frequently traversed, exhibiting high redundancy. The redundancy index for each high-frequency point in the interval attribute list is calculated, combining the average interval and periodicity: R_redundancy = P_period / (T_interval / T_max + 1), where R_redundancy is the redundancy index value (0-1), P_period is the periodicity index value (0-1), T_interval is the average interval duration in minutes, and T_max is a reference time of 20 minutes to ensure normalization of the redundancy index. The redundancy index for the high-frequency point in front of the handwashing station is approximately 0.64. A redundancy threshold is set, and high-frequency points with a redundancy index exceeding the threshold are identified as high-redundancy locations. The threshold is set to 0.6; the redundancy index of the high-frequency point in front of the washbasin is 0.64, exceeding the threshold and thus marked as high-redundant. Candidate redundancy locations are obtained based on the redundancy marking results in the interval attribute list. These candidates include all high-redundant high-frequency points and their redundancy attribute information.

[0055] Redundant location candidates are evaluated for mobility energy consumption to select retained nodes. For each redundant location candidate in the set, mobility energy consumption is evaluated to assess its impact on the overall path energy consumption. The energy consumption of the current path including the redundant location candidate and the optimized path energy consumption after removing the candidate are calculated. Energy consumption is represented as a weighted sum of the total path travel distance and the number of turns. The current path with the redundant location candidate has a total travel distance of 15 meters and 8 turns; after removing the redundant location candidate, the optimized path has a total travel distance of 13.5 meters and 6 turns. The energy consumption difference is calculated as the difference between the current energy consumption and the optimized energy consumption. A positive difference indicates that removing the redundant location candidate reduces energy consumption. Redundant location candidates with an energy consumption difference greater than a threshold of 1.0 are determined to be removable nodes, while those with an energy consumption difference less than the threshold are determined to be retained nodes. One redundant location candidate is determined to be removable because its energy consumption decreases by 1.5 units after removal. Another redundant location candidate is determined to be retained because its energy consumption only decreases by 0.3 units after removal and it increases path complexity and the risk of coverage omissions. Select and retain nodes. Retained nodes are candidate nodes that, if removed, will not significantly reduce energy consumption or will increase path complexity. These nodes need to be retained in the optimal path to ensure the integrity and effectiveness of the path.

[0056] The optimal navigation path area is formed by removing redundant nodes from the retained nodes. Retained nodes are removed from the candidate set of redundant locations, leaving the remaining candidate points as the confirmed redundant nodes. Redundant nodes are locations that can be safely removed and whose removal optimizes path efficiency; their removal does not affect the path's coverage integrity. A node removal operation is performed on the initial navigation path, removing redundant nodes from the trajectory and reconnecting adjacent retained nodes. After removing the redundant node in front of the handwashing station, the path directly connects to the ground cleaning area from the side of the handwashing station, skipping intermediate points that are repeatedly visited, making the path simpler and more efficient. The path length is shortened after the removal operation, the number of turns is reduced, and the overall movement efficiency is improved by approximately 15%. The integrity of the path after removal is verified to ensure that the path still covers all key coverage areas and that each area is reachable. The verification process is implemented through a reachability check algorithm. The reachability check verifies that each key coverage area is visited at least once in the optimized path, confirming that the path coverage reaches 100%. An optimal navigation path area is formed, which includes the optimized path node sequence, the connecting paths between nodes, the total path length, and the estimated cleaning time. The path minimizes the travel distance and energy consumption while ensuring coverage integrity.

[0057] Step S150: The robot's autonomous movement trajectory is formed through the optimal navigation path area. Based on the robot's autonomous movement trajectory, the covered area is visually detected in real time to determine the missing coverage area. The missing coverage area is compared and verified to generate supplementary coverage path control instructions.

[0058] Specifically, an autonomous robot movement trajectory is formed through an optimal navigation path area. The path node sequence and inter-node connections are extracted from the optimal navigation path area. The node sequence defines the order in which the robot should visit locations, and the connections define the movement paths between nodes. The path node sequence in the optimal navigation path area is converted into robot-executable movement commands. The conversion process calculates the movement direction, distance, and speed parameters between adjacent nodes. Moving from the washbasin location node (2.5, 3.0) to the floor cleaning node (4.0, 2.5), the movement direction angle is approximately -18 degrees, and the movement distance is approximately 1.6 meters. The robot's movement speed is set, dynamically adjusted based on the complexity of the path segments in the optimal navigation path area and the density of surrounding obstacles. The speed is set at 0.5 meters per second in open areas, 0.3 meters per second in narrow passages, and 0.2 meters per second in corner areas. A movement trajectory plan is generated, containing the start point, end point, movement direction, distance, and speed parameters for each path segment. The robot performs autonomous movement according to the trajectory plan, adjusting its position and attitude in real time to track the planned trajectory. Position feedback control is achieved through odometry and an inertial measurement unit. The robot's autonomous movement trajectory is formed, which records the actual position sequence and time sequence of the robot's movement, including the robot's coordinate position, direction of movement, speed of movement and execution status at each moment.

[0059] In some embodiments, the step of determining the missed coverage area by real-time visual detection of the covered area based on the robot's autonomous movement trajectory includes: performing cleaning coverage rate analysis on the covered area based on the robot's autonomous movement trajectory to locate the insufficient coverage area; identifying the coverage rate drop segment based on the insufficient coverage area and generating a drop marker; using the drop marker to track visual omission points and obtain a set of suspected omission locations; and combining the set of suspected omission locations to determine the missed coverage area.

[0060] Based on the robot's autonomous movement trajectory, cleaning coverage analysis is performed on the already covered area to locate under-covered areas. Position sequence data is extracted from the robot's autonomous movement trajectory, recording the robot's spatial coordinates and arrival time at each moment. The covered area corresponding to the robot's autonomous movement trajectory is calculated based on the robot's cleaning radius, which is the maximum distance the robot cleaning equipment can reach. The robot's cleaning radius is 0.25 meters; when the robot passes a certain position on its autonomous movement trajectory, a circular area with a radius of 0.25 meters centered on that position is considered the covered area. All position points on the robot's autonomous movement trajectory are traversed, and the covered area corresponding to each position point is calculated. All covered areas are then merged to form the total covered area. The total covered area is compared with the planned cleaning area, which is derived from the definition of key coverage areas and densely populated target locations. The cleaning coverage rate is calculated as the ratio of the covered area to the planned cleaning area, using the formula R_coverage = S_covered / S_planned × 100%, where R_coverage is the cleaning coverage rate in percentage, S_covered is the covered area in square meters, and S_planned is the planned cleaning area in square meters. The spatial distribution of cleaning coverage is analyzed by dividing the planned cleaning area into multiple sub-regions. The coverage rate of each sub-region is calculated, and areas with low coverage are identified. Insufficient coverage areas are located; these are sub-regions with coverage below a set threshold of 80%, requiring further coverage analysis and supplementary cleaning. The spatial boundaries, coverage rate values, and area sizes of the insufficient coverage areas are recorded.

[0061] The system identifies coverage decline segments within insufficient coverage areas and generates decline markers. It extracts the spatial boundaries and coverage data of these areas. The coverage distribution gradient within the insufficient coverage areas is analyzed to identify regions where coverage decreases from high to low. The coverage gradient is obtained by calculating the coverage difference between adjacent spatial locations within the insufficient coverage area; a negative difference indicates a decline in coverage. On the eastern side of the ground area, coverage decreases from 90% to 70% and then to 50%, forming a clear downward trend. Coverage decline segments are identified; these are continuous spatial areas where coverage continuously decreases. The criteria for identifying a decline segment are that the coverage of three or more consecutive spatial locations shows a decreasing trend with a decline exceeding 10%. On the eastern side of the ground area, the coverage from coordinates (4.0, 2.0) to (4.5, 2.0) to (5.0, 2.0) is 90%, 70%, and 50% respectively, constituting a coverage decline segment. Calculate the descent rate of the descent segment, which is the change in coverage per unit distance, using the formula V_decline = ΔR_coverage / Δd, where V_decline is the descent rate in percentage per meter, ΔR_coverage is the change in coverage in percentage, and Δd is the spatial distance in meters. On the eastern side of the ground, the coverage decreases by 40% within a 0.5-meter distance, at a rate of 80% per meter. Locate the start and end positions of the descent segment. The start position is where coverage begins to decline significantly, and the end position is where coverage reaches its lowest point or begins to recover. Generate descent markers, recording the spatial location, start and end coordinates, coverage change range, and descent rate of the descent segment, marking areas of drastic deterioration in coverage quality.

[0062] A set of suspected missed locations is obtained by visually tracking down descent markers. The area to be tracked is located based on the descent markers, specifically the descent segment marked by the markers and its surrounding buffer zone. Visual image data collected by the robot near the tracking area is retrieved; these images show the visual state of the area during and after robot cleaning. Stain detection analysis is performed on the images of the descent marker area, with the detection algorithm identifying the location and extent of residual stains. Stain detection is based on the color, texture, and contrast features of the image; stained areas show significant visual differences from the cleaned surface. The image of the eastern area shows a distinct gray stain patch of approximately 0.15 square meters near coordinates (4.8, 2.0). The stain location is compared to the robot's movement trajectory to determine if the stain area is within the robot's cleaning radius. The stain location is 0.35 meters from the nearest point the robot passed through, exceeding the cleaning radius by 0.25 meters, indicating an omission due to insufficient cleaning radius coverage. Visual missed points are identified as locations in the image where stains are detected but not covered by robot cleaning. Extract the location coordinates, stain severity, and stain area of ​​all visually overlooked points to form a set of suspected overlooked locations. The set of suspected overlooked locations includes three locations: the east side of the floor (4.8, 2.0), the side of the sink (2.3, 3.2), and the back of the toilet (3.5, 1.5), with stain severity of moderate, light, and heavy, respectively.

[0063] The suspected missed locations were aggregated and the covered missed areas were determined. Spatial cluster analysis was performed on each missed point in the suspected missed location set. The spatial distance between missed points in the suspected missed location set was calculated; missed points that were closer together might belong to the same continuous missed area. The distance between the two suspected missed locations on the east side of the ground was 0.3 meters, and the distance between the missed point on the side of the sink and other points was more than 1 meter. Missed points with a distance less than the threshold of 0.5 meters were grouped into the same cluster, which represents a spatially continuous missed area. The two suspected missed locations on the east side of the ground formed one cluster, while the missed points on the side of the sink and behind the toilet each formed their own independent cluster. The region of each cluster was expanded to include all missed points within the cluster and their surrounding buffer zone. The radius of the buffer zone was set to 0.2 meters to ensure coverage of any other undetected stains that might exist around the missed points. After expanding the cluster on the east side of the ground, a 0.6 square meter missed area was formed; after expanding the cluster on the side of the sink, a 0.2 square meter missed area was formed; and after expanding the cluster behind the toilet, a 0.3 square meter missed area was formed. All expanded clusters were merged to form a complete list covering the missed areas. The missed areas were then identified, including their spatial boundaries, area, degree of soiling, and reason for being missed. This list serves as the target area and path planning basis for supplementary cleaning tasks.

[0064] In some embodiments, the step of comparing and verifying the missing coverage area to generate supplementary coverage path control instructions includes: comparing and analyzing the missing coverage area with a standard coverage template to generate a coverage deviation mapping table; verifying and identifying blank points in the coverage deviation mapping table to generate blank identifiers; performing path filling processing based on the blank identifiers to generate a set of filled path segments; and arranging the set of filled path segments into supplementary coverage path control instructions.

[0065] A coverage deviation mapping table was generated by comparing and analyzing the missed coverage areas with the standard coverage template. The standard coverage template data was obtained, defining the ideal cleanliness and coverage requirements for each area in the restroom environment. The standard coverage template represents space in a grid format, with grid cells indicating whether they should be covered and the required coverage quality. The restroom environment was divided into 100 grid cells, of which 85 cells were marked as requiring cleaning and coverage in the standard coverage template, with a coverage quality requirement of 95% or higher. The spatial location and boundary information of the missed coverage areas were extracted and mapped to the grid system of the standard coverage template. The missed coverage areas on the east side of the ground included grid cells 45, 46, 55, and 56 (4 cells in total). The actual coverage status of the missed areas was compared with the required coverage status of the standard coverage template. Grid cell 45 had an actual coverage rate of 60%, while the standard coverage template required 95%, resulting in a deviation of -35%. Grid cell 46 had an actual coverage rate of 50%, while the standard coverage template required 95%, resulting in a deviation of -45%. Calculate the coverage deviation for each grid cell. The deviation is the difference between the actual coverage rate and the standard requirement; a negative deviation indicates insufficient coverage. Generate a coverage deviation mapping table, which records the grid cell number, standard coverage requirement, actual coverage rate, and coverage deviation for each grid cell in tabular form, visually displaying the gap between the missed areas and the standard requirements.

[0066] The coverage deviation mapping table is verified and blank points in the positioning path are identified, generating blank markers. Coverage deviation data is extracted from the coverage deviation mapping table, and grid cells with deviations exceeding a threshold are filtered. The deviation threshold is set to -20%; a deviation less than -20% indicates severe coverage deficiency in the cell. Grid cells 45 and 46, with deviations of -35% and -45% respectively, both exceed the threshold and are identified as severely under-covered cells. The severely under-covered cells in the coverage deviation mapping table are verified and identified to confirm whether the under-coverage is caused by blank areas in the robot path planning. The robot's movement trajectory data is extracted to determine whether the trajectory passes through the grid cell or its adjacent area, and the shortest distance between the trajectory and the cell center is calculated. Grid cell 45 is 0.4 meters away from the nearest trajectory point, exceeding the cleaning radius by 0.25 meters, and is identified as being caused by a blank path. Grid cell 46 is 0.38 meters away from the nearest trajectory point and is also identified as a blank path. Blank points in the path are located; these are spatial locations that should be covered but have not been reached by the robot path. The coordinates of the blank points are the center point coordinates of the under-covered cell or the internal point of the cell farthest from the trajectory. The coordinates of the blank point on the path in grid cell 45 are (4.75, 2.25), and the coordinates of the blank point on the path in cell 46 are (5.25, 2.25). Blank markers are generated, recording the coordinates of the blank point on the path, its corresponding grid cell, coverage deviation, and reason for the missing information. These markers mark key locations where path filling is needed, providing clear target points for path filling processing.

[0067] The system generates a set of filled path segments by filling in blank markers. It extracts the coordinates and reasons for any missing points on the paths marked with blank markers. A reachability analysis is performed on these blank points to determine if the robot can reach them from its current position or an existing path. Obstacle distribution and accessibility are considered to eliminate unreachable blank points. Since all blank points marked with blank markers are located within the navigable area, the reachability analysis passes. A filled path segment is planned for each blank point on the paths marked with blank markers, connecting the robot's existing path to the blank point. The path planning uses a shortest path algorithm, starting from the nearest existing path node to the blank point, while also considering path smoothness and minimizing turning times. The blank point (4.75, 2.25) is closest to the existing path node (4.0, 2.0), and the planned filled path segment moves from (4.0, 2.0) along a straight line to (4.75, 2.25), with a path length of approximately 0.8 meters. The filling path segment for the blank point (5.25, 2.25) is moved from (4.75, 2.25) to (5.25, 2.25), with a path length of 0.5 meters. The movement parameters for the filling path segment are calculated, including the start coordinates, end coordinates, movement direction, movement distance, and estimated movement time. A set of filling path segments is generated, containing the path segment data, movement parameters, and execution order corresponding to all blank points, forming a supplementary coverage path planning scheme. The set of filling path segments provides the path data foundation for the final control command generation.

[0068] The filled path segments are collected and arranged into supplementary coverage path control instructions. Spatial correlation analysis is performed on each path segment in the filled path segment set to identify the connection relationships between the path segments. Two path segments can be directly connected when their endpoint and starting point coordinates are the same or close. The endpoint (4.75, 2.25) of path segment 1 coincides with the starting point (4.75, 2.25) of path segment 2, and the two path segments in the filled path segment set can be connected sequentially. Connectable path segments are merged, and the merged path segment connects the filling paths of multiple blank points, reducing path fragmentation. Merging path segment 1 and path segment 2 forms a complete supplementary path on the east side of the ground, with the path running from (4.0, 2.0) through (4.75, 2.25) to (5.25, 2.25), with a total length of 1.3 meters. For path segments that cannot be directly connected, connecting segments are inserted, and the connecting segments plan the movement path from the endpoint of one path segment to the starting point of another path segment. The starting point (2.0, 3.0) of the filling path segment on the side of the sink is not connected to the ending point (5.25, 2.25) on the east side of the ground. A connecting segment is inserted, moving from (5.25, 2.25) to (2.0, 3.0), with a length of approximately 3.5 meters. The execution order of all path segments in the filling path segment set is arranged according to the principles of spatial proximity and path length optimization. Spatially adjacent filling path segments are executed first to reduce the robot's round-trip travel distance and improve the efficiency of supplementary cleaning. Supplementary coverage path control instructions are generated, represented in the form of a structured command sequence, including the starting point, ending point, direction of movement, speed of movement, and cleaning action parameters for each path segment. The supplementary coverage path control instructions are sent to the robot execution system, and the robot executes each filling path segment sequentially according to the instructions to complete the supplementary cleaning of missed areas, ultimately achieving complete coverage of the restroom environment.

[0069] To implement the visual recognition-based restroom cleaning path planning method corresponding to the above method embodiments, and to achieve the corresponding functional and technical effects. See also Figure 2 , Figure 2 A structural block diagram of a visual recognition-based restroom cleaning path planning system 200 provided in this application embodiment is shown. For ease of explanation, only the parts relevant to this embodiment are shown. The visual recognition-based restroom cleaning path planning system 200 provided in this application embodiment includes: The data acquisition module 201 is used to acquire multi-location visual image data of the environment in real time, extract spatial distribution information of the target area from the multi-location visual image data, and perform image segmentation processing on the spatial distribution information of the target area to form a region-location feature matrix. The region identification module 202 is used to obtain the region coverage priority classification by visual recognition based on the region-location feature matrix, to generate a coverage weight mapping library by non-linear weight allocation of the region coverage priority classification according to the region area, and to match and associate the coverage weight mapping library with the spatial distribution information of the target region to determine the key coverage area. The weight construction module 203 is used to identify the dense location of targets and the location of obstacles through the key coverage area, determine the influence weight of the surrounding navigable area based on the location of obstacles, and combine the influence weight with the dense location of targets to construct a navigable spatial weight map. The path optimization module 204 is used to obtain mobile accessibility information based on the navigable spatial weight map, identify path redundancy areas from the mobile accessibility information, and remove redundant nodes from the path redundancy areas to generate the optimal navigation path area. The execution control module 205 is used to form the robot's autonomous movement trajectory through the optimal navigation path area, perform real-time visual detection on the covered area based on the robot's autonomous movement trajectory to determine the covered omission area, and compare and verify the covered omission area to generate supplementary coverage path control instructions.

[0070] The aforementioned visual recognition-based restroom cleaning path planning system 200 can implement the visual recognition-based restroom cleaning path planning method of the above method embodiments. The options in the above method embodiments are also applicable to this embodiment, and will not be detailed here. The remaining content of this application embodiment can be referred to the content of the above method embodiments, and will not be repeated in this embodiment.

[0071] The purpose of the above embodiments is to reproduce and derive the technical solution of the present invention by way of example, and to fully describe the technical solution, purpose and effect of the present invention. The purpose is to enable the public to have a more thorough and comprehensive understanding of the disclosure of the present invention, and not to limit the scope of protection of the present invention.

[0072] The above embodiments are not an exhaustive list based on the present invention, and there may be many other embodiments not listed. Any substitutions and improvements made without departing from the concept of the present invention are within the protection scope of the present invention.

Claims

1. A method for planning toilet cleaning paths based on visual recognition, characterized in that, include: Real-time acquisition of multi-location visual image data of the environment; extraction of target region spatial distribution information from the multi-location visual image data; and image segmentation processing of the target region spatial distribution information to form a region-location feature matrix. Based on the region-location feature matrix, visual recognition is performed to obtain the region coverage priority classification. The region coverage priority classification is non-linearly weighted according to the region area to generate a coverage weight mapping library. The coverage weight mapping library is matched and associated with the spatial distribution information of the target region to determine the key coverage area. The key coverage area is used to identify dense target locations and obstacle occlusion locations. The influence weight of the surrounding navigable area is determined based on the obstacle occlusion locations. The influence weight is combined with the dense target locations to construct a navigable spatial weight map. Based on the navigable spatial weight map, mobile accessibility information is obtained, and redundant paths are identified from the mobile accessibility information. Redundant nodes in the redundant paths are removed to generate the optimal navigation path area. The robot's autonomous movement trajectory is formed through the optimal navigation path area. Based on the robot's autonomous movement trajectory, real-time visual detection is performed on the covered area to determine the missed coverage area. The missed coverage area is compared and verified to generate supplementary coverage path control instructions.

2. The method according to claim 1, characterized in that, The step of generating a coverage weight mapping library by non-linearly weighting the regional coverage priority classification according to the regional area includes: The area coverage priority is classified and grouped by area size to obtain the area classification results; Based on the area grading results, the differences in cleaning difficulty are identified, and the difficulty grading points are determined. The difficulty level points are converted into weight gain factors by threshold segmentation mapping to form a factor allocation table; A coverage weight mapping library is established based on the aforementioned factor allocation table.

3. The method according to claim 1, characterized in that, The determination of the influence weight of the surrounding navigable area based on the obstacle occlusion position includes: The areas where obstacles obstruct the view are divided into fixed facility areas and moving object areas; The cleaning radius detection signal is transmitted from the fixed facility area to the moving object area; The avoidance distance data of the cleaning radius detection signal is obtained to form a distance parameter table; Based on the distance parameter table, the area with the largest clean reach is selected as the influence weight of the surrounding navigable area.

4. The method according to claim 1, characterized in that, The step of combining the influence weights with the dense locations of the targets to construct a navigable spatial weight map includes: Based on the analysis of the densely located targets, a stain aggregation characteristic is constructed to create an aggregation relationship diagram. The spatial proximity of the clustering graph is determined to obtain a proximity data table; The sparse region synergy effect of the sparse region is obtained by performing sparse region mining on the proximity data table. The sparse region synergy effect and the influence weight are fused to form a navigable spatial weight map.

5. The method according to claim 1, characterized in that, The step of removing redundant nodes from the redundant path region to generate the optimal navigation path region includes: The redundant areas of the path are subjected to repeatability detection to obtain round-trip movement records; Identify repeatedly visited points from the round-trip travel records as redundant location candidates; The redundant location candidates are evaluated for mobile energy consumption to filter and retain nodes; The optimal navigation path area is formed by removing the remaining redundant nodes based on the retained nodes.

6. The method according to claim 1, characterized in that, The step of determining missed coverage areas by real-time visual detection of the covered area based on the robot's autonomous movement trajectory includes: Based on the robot's autonomous movement trajectory, the cleaning coverage rate of the covered area is analyzed to locate areas with insufficient coverage. Based on the insufficient coverage area, identify the coverage drop segment and generate a drop marker; The set of suspected missed locations is obtained by visually tracking the descent markers. The suspected missing locations are collected and the area to be covered is determined.

7. The method according to claim 1, characterized in that, The step of comparing and verifying the missing coverage areas to generate supplementary coverage path control instructions includes: A coverage deviation mapping table is generated by comparing and analyzing the missing coverage areas with the standard coverage template. The coverage deviation mapping table is verified to identify blank points in the positioning path and generate blank markers. Based on the blank identifier, a path filling process is performed to generate a set of filled path segments; The set of filled path segments is arranged into supplementary overlay path control instructions.

8. The method according to claim 3, characterized in that, The process of obtaining the avoidance distance data from the clean radius detection signal to form a distance parameter table includes: Identify the detection termination position from the cleaning radius detection signal; The detection termination position is segmented and segmented to generate segmented distance groups; Based on the segmented distance groups, periodic repeating attributes are extracted to obtain a repeating pattern table; The repetition pattern table is combined with the detection termination position to generate a distance parameter table by distance classification and marking.

9. The method according to claim 5, characterized in that, The step of identifying repeatedly visited points as redundant location candidates from the round-trip movement records includes: Frequency statistics are obtained by performing frequency detection on the round-trip movement records; High-frequency points are identified from the frequency statistics results to generate a high-frequency point set; Extract the repetition interval attribute from the high-frequency point set to form an interval attribute list; Candidate redundant positions are obtained by marking the degree of redundancy based on the interval attribute list.

10. A visual recognition-based restroom cleaning path planning system, characterized in that, include: The data acquisition module is used to acquire multi-location visual image data of the environment in real time, extract spatial distribution information of the target area from the multi-location visual image data, and perform image segmentation processing on the spatial distribution information of the target area to form a region-location feature matrix. The region identification module is used to obtain the region coverage priority classification by visual recognition based on the region-location feature matrix, to generate a coverage weight mapping library by non-linear weight allocation of the region coverage priority classification according to the region area, and to match and associate the coverage weight mapping library with the spatial distribution information of the target region to determine the key coverage area. The weight construction module is used to identify the dense locations of targets and the locations of obstacles through the key coverage area, determine the influence weight of the surrounding navigable area based on the location of obstacles, and combine the influence weight with the dense locations of targets to construct a navigable spatial weight map. The path optimization module is used to obtain mobile accessibility information based on the navigable spatial weight map, identify redundant areas in the path from the mobile accessibility information, and remove redundant nodes in the redundant areas to generate the optimal navigation path area. The execution control module is used to form the robot's autonomous movement trajectory through the optimal navigation path area, perform real-time visual detection on the covered area based on the robot's autonomous movement trajectory to determine the covered areas that are missing, and compare and verify the covered areas to generate supplementary coverage path control instructions.

Images