Environment construction method, electronic equipment, sweeping robot and computer program product

Abstract

The invention provides an environment construction method, electronic equipment, a sweeping robot and a computer program product. The environment construction method comprises the steps that local observation data, collected by a sweeping robot, of different environment attributes are acquired for a to-be-cleaned area; preprocessing the local observation data of different environment attributes, and determining environment attribute observation vectors of different environment attributes; mapping the environment attribute observation vectors to corresponding grid units of a global grid map, and determining environment attribute potential field matrixes of different environment attributes; based on the environment attribute observation vectors of different environment attributes, performing recursive fusion on the environment attribute states of the grid units in the corresponding environment attribute potential field matrixes, and determining a basic environment attribute potential field; and spatial diffusion processing is carried out on the basic environment attribute potential field, and target environment attribute potential fields of different environment attributes are determined.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to an environment construction method, an electronic device, a sweeping robot and a computer program product. BACKGROUND

[0002] The environment perception and task planning of the existing sweeping robot mainly rely on a geometric map, and obstacles and passable areas are identified through a laser radar or a visual sensor to realize basic path coverage and obstacle avoidance functions. Only the spatial structure is concerned, and the semantic properties of the environment are not deeply understood, and historical observations and real-time observation data are not effectively fused, resulting in unstable evaluation of contaminated areas and susceptibility to noise interference, and complex phenomena such as stain diffusion or residue cannot be simulated. SUMMARY

[0003] The present disclosure provides an environment construction method, an electronic device, a sweeping robot and a computer program product.

[0004] According to one aspect of the present disclosure, an environment construction method is provided, comprising: for a to-be-cleaned area, acquiring local observation data of different environment attributes collected by a sweeping robot, the environment attributes including ground material, ground dirtiness, ground stain distribution and / or odor concentration; pre-processing the local observation data of different environment attributes to determine environment attribute observation vectors of different environment attributes, the pre-processing including feature extraction, normalization processing and confidence calculation; mapping the environment attribute observation vectors to corresponding grid cells of a global grid map to determine environment attribute potential field matrices of different environment attributes, the global grid map being obtained by grid division on the to-be-cleaned area; based on the environment attribute observation vectors of different environment attributes, respectively performing recursive fusion on the environment attribute states of the grid cells in the corresponding environment attribute potential field matrices to determine a basic environment attribute potential field, the environment attribute states being determined based on historical environment attribute observation vectors; and respectively performing spatial diffusion processing on the basic environment attribute potential field to determine target environment attribute potential fields of different environment attributes.

[0005] According to the environment construction method of one aspect, the local observation data is subjected to feature extraction, normalization and confidence evaluation, and is dynamically mapped to a global grid map, ensuring spatial alignment and reliability quantification of the perception information. Based on the recursive fusion mechanism, the environment attribute states of each grid cell are continuously updated to realize online tracking and uncertainty management of environment changes, and a robust basic environment attribute potential field is formed. Further through spatial diffusion processing, the natural propagation law of pollutants is simulated, the observation blind area of the sensor is compensated, and the spatial continuity and structural consistency of the potential field are enhanced. The finally generated multi-attribute target environment attribute potential field not only accurately reflects the current pollution distribution, but also has physical interpretability and spatio-temporal evolution ability.

[0006] According to at least one embodiment of the environment construction method of this disclosure, based on the environmental attribute observation vectors of different environmental attributes, the environmental attribute states of the grid cells in the corresponding environmental attribute potential field matrix are recursively fused, including: when the environmental attribute is the degree of ground dirtiness, at each time step, the observation probability in the corresponding environmental attribute observation vector is converted into a log-probability increment; the log-probability increment is recursively fused with the environmental attribute states of the grid cells in the environmental attribute potential field matrix to determine the basic environmental attribute potential field.

[0007] According to at least one embodiment of the environment construction method of this disclosure, spatial diffusion processing is performed on the basic environmental attribute potential field to determine the target environmental attribute potential field for different environmental attributes, including: determining the observation confidence of each grid cell based on local observation data collected at the location of each grid cell in the basic environmental attribute potential field; determining the spatial diffusion intensity based on a gating function constructed based on the observation confidence; calculating the gradient magnitude of the basic environmental attribute potential field to determine the spatial diffusion boundary; and performing spatial diffusion processing on the basic environmental attribute potential field based on the spatial diffusion intensity and the spatial diffusion boundary to determine the target environmental attribute potential field for different environmental attributes.

[0008] According to at least one embodiment of the environment construction method of this disclosure, the basic environmental attribute potential field is subjected to spatial diffusion processing to determine the target environmental attribute potential field for different environmental attributes, including: taking grid cells as nodes, connecting adjacent grid cells through edges, determining edge weights based on the physical distance between grid cells, the direction of pedestrian flow and the continuity of ground material, and determining the adjacency graph of the basic environmental attribute potential field; and performing spatial diffusion processing based on the adjacency graph to determine the target environmental attribute potential field for different environmental attributes.

[0009] According to at least one embodiment of the environment construction method of this disclosure, after determining the target environmental attribute potential field with different environmental attributes, the method includes: real-time acquisition of the motor current change, suction fluctuation and visual frame difference of the sweeping robot; weighted fusion of the motor current change, suction fluctuation and visual frame difference through a nonlinear regression model to determine the cleaning success probability; and updating the environmental attribute state of the grid cells in the target environmental attribute potential field based on the cleaning success probability.

[0010] According to at least one embodiment of the environmental construction method of this disclosure, after determining the target environmental attribute potential field with different environmental attributes, the method includes: constructing a residual kernel function based on the characteristics of the ground material, the movement direction of the sweeping robot, and the cleaning mode; calculating the amount of pollutant residue for the target grid cell whose cleaning success probability is less than the target probability based on the cleaning success probability and the environmental attribute state; and weighting the amount of pollutant residue with the residual kernel function to update the environmental attribute state of the adjacent grid cells of the target grid cell.

[0011] According to at least one embodiment of the environment construction method of this disclosure, after determining the target environmental attribute potential field with different environmental attributes, the method includes: performing a weighted calculation on the environmental attribute state of each grid cell in the target environmental attribute potential field with different environmental attributes to determine the cleaning priority index of each grid cell in the target environmental attribute potential field; and performing path planning for the cleaning task of the sweeping robot according to the cleaning priority index using a path planning algorithm.

[0012] According to another aspect of this disclosure, an electronic device is provided, comprising: a memory storing execution instructions; and a processor executing the execution instructions stored in the memory, causing the processor to perform an environment construction method according to any embodiment of this disclosure.

[0013] According to another aspect of this disclosure, a robotic vacuum cleaner is provided, the robotic vacuum cleaner including the above-described electronic device.

[0014] According to another aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements an environment construction method according to any embodiment of this disclosure. Attached Figure Description

[0015] The accompanying drawings illustrate exemplary embodiments of the present disclosure and, together with the description thereof, serve to explain the principles of the present disclosure. These drawings are included to provide a further understanding of the present disclosure and are incorporated in and constitute a part of this specification.

[0016] Figure 1 This is an overall flowchart of an environment construction method according to one embodiment of the present disclosure.

[0017] Figure 2 This is a flowchart illustrating the process of determining the basic environmental attribute potential field in an environment construction method according to one embodiment of the present disclosure.

[0018] Figure 3 This is a flowchart illustrating the process of determining the target environment attribute potential field in an environment construction method according to one embodiment of the present disclosure.

[0019] Figure 4 This is a flowchart illustrating the process of determining the target environment attribute potential field in an environment construction method according to another embodiment of the present disclosure.

[0020] Figure 5 This is a schematic diagram of the process of updating the environment attribute state in an environment construction method according to one embodiment of the present disclosure.

[0021] Figure 6 This is a schematic flowchart illustrating the process of updating the environmental attribute state of adjacent grid cells of a target grid cell in an environment construction method according to one embodiment of the present disclosure.

[0022] Figure 7 This is a flowchart illustrating the path planning for the cleaning task of a robotic vacuum cleaner in an environment construction method according to one embodiment of the present disclosure.

[0023] Figure 8 This is a schematic structural block diagram of an environment construction apparatus according to one embodiment of the present disclosure.

[0024] Figure 9 This is a schematic structural block diagram of an electronic device according to one embodiment of the present disclosure. Detailed Implementation

[0025] The present disclosure will now be described in further detail with reference to the accompanying drawings and examples. It should be understood that the specific examples described herein are for illustrative purposes only and are not intended to limit the scope of the disclosure. Furthermore, it should be noted that, for ease of description, only the parts relevant to the present disclosure are shown in the accompanying drawings.

[0026] It should be noted that, where there is no conflict, the embodiments and features described in this disclosure can be combined with each other. The technical solutions of this disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0027] When there is a mixture of dried coffee stains and ordinary dust on the ground, the robot vacuum cleaner has difficulty distinguishing the differences in their properties and often cleans with the same suction mode, resulting in stubborn stains remaining while the dusty areas are over-cleaned. In the space where the kitchen and living room are connected, as oil stains spread with the flow of people, traditional path planning based on geometric coverage cannot perceive this semantic distribution of dirt, resulting in unreasonable allocation of cleaning resources.

[0028] To address this, this disclosure proposes an environment construction method. By integrating multimodal data from vision, force sensing, and gas sensing, it constructs multidimensional attribute potential fields for ground dirt levels, stain distribution density, and odor concentration. This effectively distinguishes between highly adhesive, stubborn stains and easily removable dust. Furthermore, it dynamically adjusts cleaning strategies based on ground material characteristics, employing a intensive wet mopping mode for high-risk stain areas and efficient, low-consumption sweeping for dusty areas to avoid resource waste. Simultaneously, a spatial diffusion mechanism expands local observations into a semantic pollution propagation map, accurately capturing the trend of stain spread and overcoming the limitations of traditional geometric coverage path planning. By recursively fusing historical event frequencies with real-time feedback, it further identifies high-frequency pollution zones and prioritizes them, enabling the robot vacuum to proactively focus on critical paths of pollution evolution.

[0029] To facilitate description and make the technical solutions of this disclosure easier to understand, the terminology of this disclosure will be explained before describing the technical solutions of this disclosure.

[0030] A potential field is the environment in which objects interact in space. It is a field of energy distribution. Through each point in the potential field, a specific potential value can be determined, that is, the state of the environmental properties.

[0031] The environment construction method disclosed herein can be deployed on embedded processors of terminal devices such as robotic vacuum cleaners to achieve fully localized real-time environmental perception, potential field updates, and path planning, ensuring reliable operation in scenarios without network access or with high privacy requirements, such as homes or offices. It can also be deployed on cloud servers to receive local observation data from multiple robotic vacuum cleaners, performing large-scale cross-regional joint potential field calculations and long-term evolution analysis, suitable for collaborative scheduling and cleaning strategy optimization of multiple robotic vacuum cleaners in large venues such as shopping malls or airports.

[0032] Figure 1 A schematic diagram illustrating the overall flow of an environment construction method according to one embodiment of this disclosure is shown. Figure 1 The method M100 shown includes steps S110 to S140. This method can be performed by a robotic vacuum cleaner.

[0033] In step S110, for the area to be cleaned, local observation data of different environmental attributes collected by the sweeping robot are obtained. The environmental attributes include the ground material, the degree of dirt on the ground, the distribution of stains on the ground and / or the concentration of odors.

[0034] The area to be cleaned is the physical space range in which the robot vacuum cleaner needs to perform cleaning tasks. It is usually represented by a global map constructed using synchronous localization and mapping technology, and further divided into a global grid composed of several regular grids.

[0035] Preferably, the image recognition module of the sweeping robot detects the type and area of ​​stains on the ground to determine the distribution of stains; the odor detection module judges the odor concentration distribution to determine the odor concentration; the infrared sensor judges the reflectivity and thermal characteristics of the ground material to determine the ground material; and the friction feedback module judges the roughness and humidity of the ground to determine the degree of dirtiness.

[0036] In step S120, the local observation data of different environmental attributes are preprocessed to determine the environmental attribute observation vectors of different environmental attributes. The preprocessing includes feature extraction, normalization and confidence calculation.

[0037] Feature extraction is performed on local observation data for each type of environmental attribute to identify key discriminant indicators. Numerical normalization is then performed to eliminate dimensional differences, and the confidence score of the current local observation data is calculated to determine the structured environmental attribute observation vector.

[0038] Optionally, confidence levels can be calculated for local observation data based on the sensor operating status of the robotic vacuum cleaner, environmental interference, and sampling time interval.

[0039] In step S130, the environmental attribute observation vector is mapped to the corresponding grid cell of the global grid map to determine the environmental attribute potential field matrix of different environmental attributes. The global grid map is obtained by dividing the area to be cleaned into grids.

[0040] For each type of environmental attribute, maintain an environmental attribute potential field matrix that is isomorphic to the global raster map. Each element in the environmental attribute potential field matrix represents the current observation or initial estimate of the corresponding raster cell on that environmental attribute, i.e., the environmental attribute state.

[0041] In step S140, based on the environmental attribute observation vectors of different environmental attributes, the environmental attribute states of the grid cells in the corresponding environmental attribute potential field matrix are recursively fused to determine the basic environmental attribute potential field. The environmental attribute states are determined based on the historical environmental attribute observation vectors.

[0042] For each type of environmental attribute potential field matrix, the newly acquired environmental attribute observation vector is recursively fused with the environmental attribute state of its corresponding grid cell, using grid cells as units. Instead of relying solely on the environmental attribute state as sufficient statistics to encode past information, the current best estimate is continuously updated using Bayesian filtering, ultimately generating a basic environmental attribute potential field that reflects long-term accumulated knowledge.

[0043] Preferably, the environmental attribute state is a quantized value, i.e., a potential field value, used to characterize the grid cell under a specific environmental attribute. A dynamically robust environmental attribute state is formed by fusing multi-source historical local observation data with the latest local observation data through a recursive Bayesian framework.

[0044] Preferably, for environmental attribute observation vectors whose environmental attributes are ground stain distribution or odor concentration, the environmental attribute states of the corresponding environmental attribute potential field matrix grid cells are recursively fused using Dirichlet parameters to determine the basic environmental attribute potential field.

[0045] Preferably, for environmental attribute observation vectors whose environmental attribute is ground material, the environmental attribute state of the corresponding environmental attribute potential field matrix grid cells is recursively fused using Dirichlet parameters to determine the basic environmental attribute potential field.

[0046] In step S150, the basic environmental attribute potential fields are subjected to spatial diffusion processing to determine the target environmental attribute potential fields for different environmental attributes.

[0047] Using the potential field value (i.e., environmental attribute state) of the current grid cell as the source, and based on a predefined diffusion kernel function, its influence is transferred to the neighboring region of the current grid cell with a certain weight. This process simulates the natural spread effect of pollutants in the real environment caused by factors such as human movement, airflow, or material continuity, generating a target environmental attribute potential field with greater spatial consistency and physical rationality.

[0048] Preferably, the potential field of the basic environmental attribute is spatially diffused by multi-scale convolutional diffusion, confidence-gated anisotropic diffusion, and graph Laplace directional diffusion to determine the potential field of the target environmental attribute for different environmental attributes.

[0049] Therefore, the environmental construction method disclosed herein is based on multimodal sensors such as image, infrared, odor, and friction feedback to collect local observation data such as ground material, degree of dirt, stain distribution, and odor concentration. Data preprocessing is completed through feature extraction, normalization, and confidence assessment. Structured observation vectors are mapped to a global grid map, and the state estimates of each attribute are continuously updated using a Bayesian recursive fusion mechanism to form a basic environmental attribute potential field with time memory capabilities. Furthermore, spatial diffusion processing is introduced, combining material continuity, pedestrian flow direction, and diffusion kernel function to simulate the natural propagation law of pollutants, compensating for perception blind spots and enhancing the spatial consistency of the map. The resulting multidimensional target environmental attribute potential field can not only distinguish between stubborn stains and floating dust and identify the trend of kitchen grease spreading to the living room, but also support dynamic adjustment of cleaning strategies, improving the robot's cognitive accuracy, cleaning priority judgment ability, and overall operational intelligence level in real dynamic environments.

[0050] Regarding step S140, based on the environmental attribute observation vectors of different environmental attributes, the environmental attribute states of the grid cells in the corresponding environmental attribute potential field matrix are recursively fused to determine the basic environmental attribute potential field. In some embodiments of this disclosure, this may include, for example: Figure 2Steps S1401 to S1402 are shown.

[0051] In step S1401, when the environmental attribute is the degree of ground dirtiness, at each time step, the observation probability in the corresponding environmental attribute observation vector is converted into a log-probability increment.

[0052] When the environmental attribute is the degree of ground dirtiness, at each time step, the observation probability of the environmental attribute observation vector corresponding to the current grid cell is first obtained; the observation probability is mapped to the log probability space, and the natural logarithm of the ratio of the observation probability to the complement probability is calculated to obtain the increment of the corresponding log likelihood ratio, thereby converting the probability value into an evidence strength signal in the real number domain.

[0053] From the environmental attribute observation vector collected and generated at the current moment, the observation probability of the environmental attribute being the degree of ground dirtiness is extracted. This observation probability value is mapped to the log-probability space, such that the closer the observation probability is to 1, the larger the positive increment of the log-probability, and the closer the observation probability is to 0, the larger the negative increment of the log-probability, thus transforming the observation information into an evidence weight form suitable for Bayesian recursive updates.

[0054] In step S1402, the logarithmic probability increment is recursively fused with the environmental attribute state of the grid cell in the environmental attribute potential field matrix to determine the basic environmental attribute potential field.

[0055] The logarithmic probability increment is weighted and accumulated with the environmental attribute state of the corresponding grid cell in the environmental attribute potential field matrix. By fusing the logarithmic probability increment, the environmental attribute state is continuously updated, ultimately generating a basic environmental attribute potential field that reflects long-term accumulated knowledge.

[0056] Preferably, when the environmental attribute is the degree of ground dirtiness, the environmental attribute state is: This represents the environmental attribute state at time t. This represents the environmental attribute state at time t-1. This represents the logarithmic probability increment. Indicates the probability of observation. denoted as the adaptive gain factor, and c represents the observation confidence of the current grid cell.

[0057] Therefore, converting the currently perceived observation probability into a log-probability increment and mapping it to an additivity evidence strength signal effectively solves the problem of the difficulty in directly fusing the original probability values. By recursively fusing the log-probability increment with historical environmental attribute states, and dynamically updating the pollution belief of the grid cells using Bayesian filtering, it can quickly respond to newly emerging dirty areas while suppressing misjudgments caused by sensor noise. Furthermore, by introducing an adaptive gain factor and observation confidence, and dynamically adjusting the update step size based on observation quality, map stability is maintained while ensuring response sensitivity. The resulting basic environmental attribute potential field not only reflects the current observation results but also incorporates long-term spatiotemporal evolution information, possessing both memory capacity and uncertainty expression capabilities.

[0058] Regarding step S150, which involves performing spatial diffusion processing on the basic environmental attribute potential fields to determine the target environmental attribute potential fields for different environmental attributes, some embodiments of this disclosure may include, for example... Figure 3 Steps S1501 to S1504 are shown.

[0059] In step S1501, the observation confidence of each grid cell is determined based on the local observation data collected at the location of each grid cell in the basic environmental attribute potential field.

[0060] Based on the position of each grid cell in the basic environmental potential field, multimodal sensor data corresponding to the current moment is extracted, including image clarity, illumination intensity, sensor operating status, and environmental interference information. The reliability of this observation is evaluated based on the stability of texture features output by the image recognition module, edge contrast, and motor load fluctuations of the friction feedback module. The presence of motion blur or sampling inaccuracies is determined by combining the robot's pose accuracy and the flatness of the ground. Finally, the observation confidence score of the grid cell is generated by integrating the above factors using preset rules or a lightweight classification model.

[0061] In step S1502, the spatial diffusion intensity is determined based on the gating function constructed from the observation confidence level.

[0062] Using the observation confidence of each grid cell as input, a gating weight is generated by setting a threshold or employing a sigmoid function for nonlinear mapping. This gating weight is used to adjust the propagation intensity during the spatial diffusion process. When the observation confidence is high, the gating weight approaches 1, allowing the underlying environmental property potential field to diffuse sufficiently within the grid cell and its neighborhood, thus enhancing semantic continuity. When the observation confidence is low, the gating weight approaches 0, suppressing diffusion behavior and preventing false propagation caused by unreliable observations. Multiplying the gating weight by the diffusion kernel function achieves differentiated spatial diffusion intensity modulation for different regions.

[0063] In step S1503, the gradient magnitude of the basic environmental property potential field is calculated to determine the spatial diffusion boundary.

[0064] Based on the spatial distribution of the basic environmental attribute potential field, the differential gradients of the basic environmental attribute potential field in the horizontal and vertical directions are calculated cell by cell to obtain the gradient vector of each cell. The Euclidean norm of this gradient vector is obtained as the gradient magnitude, reflecting the intensity of local changes. The gradient magnitude is binarized or weighted to identify high-gradient regions as potential spatial diffusion boundaries. Diffusion behavior is suppressed at these locations to maintain the clarity of the pollution area edge, thereby determining the boundary constraints for anisotropic diffusion control.

[0065] In step S1504, based on the spatial diffusion intensity and spatial diffusion boundary, the basic environmental attribute potential field is subjected to spatial diffusion processing to determine the target environmental attribute potential field for different environmental attributes.

[0066] The spatial diffusion intensity of each grid cell is determined by a gating function constructed based on the observation confidence level, which is used to adjust the propagation activity during the diffusion process. The spatial diffusion boundary calculated by combining the gradient magnitude is used to suppress diffusion in high gradient regions and enhance smoothness in flat regions. An anisotropic diffusion model is adopted, in which the diffusion intensity and boundary information are jointly applied to the diffusion coefficient to control the potential field to diffuse tangentially along the edge while suppressing normal penetration. The environmental attribute state values ​​of each grid cell are updated iteratively in rounds to generate a target environmental attribute potential field with different environmental attributes that retains structural features and has spatial continuity.

[0067] Regarding step S150, which involves performing spatial diffusion processing on the basic environmental attribute potential field to determine the target environmental attribute potential field for different environmental attributes, in other embodiments of this disclosure, it may include, as follows: Figure 4 Steps S410 to S420 are shown.

[0068] In step S410, grid cells are used as nodes, and adjacent grid cells are connected by edges. The edge weights are determined by the physical distance between grid cells, the direction of pedestrian flow, and the continuity of ground material, thus determining the adjacency graph of the basic environmental attribute potential field.

[0069] Preferably, the adjacency graph is represented as follows: , Represents an adjacency graph. Represents a set of nodes. Denotes the set of edges. Represents edge weight, This represents the edge weight from node i to node j. and These represent the environmental attribute observation vectors of the i-th and j-th raster cells, respectively. Represents the Gaussian kernel function. Indicates bandwidth parameter, and This represents the dominant direction vectors of the i-th and j-th grid cells. This represents the directional enhancement coefficient.

[0070] In step S420, spatial diffusion processing is performed based on the adjacency graph to determine the target environmental attribute potential field for different environmental attributes.

[0071] The diffusion flux of each node in the adjacency graph is calculated using the graph Laplacian operator. In each iteration, the environmental attribute state of the current node is dynamically adjusted based on the state differences and connection weights of neighboring nodes, so that the high pollution belief is preferentially propagated along the high weight path. The adjacency graph diffusion process is repeated until the target number of iterations is reached, generating a target environmental attribute potential field with different environmental attributes that has spatial continuity and structural preservation.

[0072] Therefore, by constraining edge weights with physical distance, pedestrian flow direction, and ground material continuity, the graph model can accurately reflect the actual patterns of pollutant spread along high-frequency paths and similar regions. During the diffusion process, the graph Laplacian operator is used to guide high-pollution beliefs to preferentially propagate along paths with structural consistency and directional coherence. This enhances the spatial continuity of the potential field, effectively preserves the true boundaries, and avoids false cross-regional diffusion. The generated target environmental attribute potential field not only integrates local observation information but also introduces global semantic priors.

[0073] In some embodiments of this disclosure, after determining the target environmental attribute potential field with different environmental attributes, the following can be included: Figure 5 Steps S510 to S530 are shown.

[0074] In step S510, the changes in motor current, suction fluctuations, and visual frame differences of the sweeping robot are collected in real time.

[0075] The robot vacuum cleaner uses its built-in sensors to synchronously acquire current signals from the main brush motor and side brush motor at a fixed sampling frequency, recording their dynamic changes and calculating the deviation from the no-load reference value. Simultaneously, it collects real-time suction data, detecting the amplitude and duration of suction fluctuations. Furthermore, it continuously captures image streams, calculating pixel-level differences between adjacent visual frames frame by frame, extracting motion area and texture change information, and obtaining the robot vacuum cleaner's motor current changes, suction fluctuations, and visual frame differences.

[0076] In step S520, the changes in motor current, suction fluctuations, and visual frame differences are weighted and fused using a nonlinear regression model to determine the probability of successful cleaning.

[0077] The changes in motor current, suction fluctuations, and visual frame differences are fused using a nonlinear regression model, such as a neural network or a multi-class decision tree. The nonlinear regression model calculates the success probability of the current cleaning task based on the learned patterns and parameters.

[0078] In step S530, the environmental attribute state of the grid cells in the target environmental attribute potential field is updated based on the success probability of cleaning.

[0079] The success rate of cleaning is converted into a log-probability increment. Based on the robot's current location, the corresponding grid cell in the target environmental attribute potential field is located, and its original environmental attribute state is obtained. The log-probability increment is then weighted and fused with the original environmental attribute state. If the success rate of cleaning is high, the environmental attribute state value of that grid cell is reduced; otherwise, the original environmental attribute state value is maintained or increased, ensuring that the target environmental attribute potential field dynamically reflects the actual cleaning effect.

[0080] Therefore, by integrating multimodal real-time sensing data such as motor current changes, suction fluctuations, and visual frame differences, a nonlinear regression model is used to accurately estimate the cleaning success probability, achieving intelligent judgment of cleaning effectiveness. Based on the cleaning success probability, the environmental attribute state of the corresponding grid cell in the target environmental attribute potential field is updated in a closed loop. This effectively improves the timeliness and accuracy of the global map state, enabling the robot vacuum to dynamically identify cleaned areas and reduce their contamination perception, avoiding repeated cleaning. At the same time, when cleaning fails, the contamination characterization is retained or enhanced, ensuring that missed areas are continuously monitored, thus improving cleaning efficiency and global coverage integrity.

[0081] In some embodiments of this disclosure, after determining the target environmental attribute potential field with different environmental attributes, the following can be included: Figure 6 Steps S610 to S630 are shown.

[0082] In step S610, a residual kernel function is constructed based on the characteristics of the ground material, the movement direction of the sweeping robot, and the cleaning mode.

[0083] Based on the characteristics of the floor material, the movement direction of the robotic vacuum cleaner, and the cleaning mode, this study analyzes the impact of different floor materials on pollutant residue and determines the material-related basic weights for residue. A directional consistency factor is constructed by combining the current movement direction of the robotic vacuum cleaner with its historical trajectory distribution to reflect the tendency for pollutants to accumulate more easily along the main path. Corresponding cleaning efficiency attenuation coefficients are set according to the cleaning mode (e.g., powerful vacuuming, gentle sweeping, or corner cleaning). The three factors are then weighted and fused to obtain the residue kernel function.

[0084] Preferably, the residual kernel function is The residual kernel function represents the kernel offset relative to the current raster cell. The relative intensity of pollutants that may remain at a given location due to incomplete cleaning; This represents the isotropic Gaussian attenuation term. Represents the offset vector. Indicates the dominant direction vector. This represents the directional modulation gain coefficient.

[0085] In step S620, for target grid cells where the success rate of cleaning is less than the target probability, the amount of pollutant residue is calculated based on the success rate of cleaning and the environmental attribute status.

[0086] For target grid cells with a cleaning success probability less than the target probability, the cleaning success probability is used as a discount factor for the cleaning effect based on the environmental attribute state of the target grid cell at the current moment, and the amount of pollutants that could not be successfully removed is calculated.

[0087] In step S630, the pollutant residue amount and the residue kernel function are weighted and allocated to update the environmental attribute status of the adjacent grid cells of the target grid cell.

[0088] Centered on the target grid cell, the residual pollutant amount is weighted for diffusion within its neighborhood based on the spatial weight distribution of the residual kernel function, and the residual contribution value borne by each adjacent grid cell is calculated. The residual contribution value is converted into a log-probability increment and weighted and fused with the current environmental attribute state of the corresponding adjacent grid cells. This enables the reasonable spread of residual pollution to the surrounding area after cleaning failure, dynamically corrects environmental perception, and ensures that the impact range of incompletely cleaned areas is accurately modeled.

[0089] Therefore, it effectively enhances the closed-loop correction capability of environmental cognition, avoids map misjudgment caused by a single cleaning failure, and enhances the spatial rationality and physical authenticity of pollution belief updates in typical scenarios such as carpets easily accumulating dirt or continuous pollution along main pedestrian paths.

[0090] In some embodiments of this disclosure, after determining the target environmental attribute potential field with different environmental attributes, the following can be included: Figure 7 Steps S710 to S720 are shown.

[0091] In step S710, the environmental attribute states of each grid cell in the target environmental attribute potential field with different environmental attributes are weighted and calculated to determine the cleaning priority index of each grid cell in the target environmental attribute potential field.

[0092] When performing weighted calculations on the environmental attribute states of each grid cell in the target environmental attribute potential field with different environmental attributes, the environmental attribute state values ​​of each grid cell under various environmental attributes are extracted, and corresponding weight coefficients are assigned according to the importance of each environmental attribute state value to the cleaning decision. The comprehensive cleaning priority index of the grid cell is obtained by multiplying the environmental attribute state values ​​at the same location by their corresponding weights and then summing the results.

[0093] Preferably, environmental attribute status values ​​with high pollution intensity, high confidence, and originating from key areas (such as the living room and kitchen) are given higher weight.

[0094] In step S720, the cleaning task of the robot vacuum cleaner is planned according to the cleaning priority index and the path planning algorithm.

[0095] Based on the cleaning priority index, the priority value of each grid cell is determined. The priority value of each grid cell is then converted into a cost map for path planning, where high-priority areas correspond to low passage costs. A global path planning algorithm (such as A) is then used. By combining Dijkstra's algorithm with local optimization strategies, an optimal cleaning trajectory is generated that covers high-priority areas, reduces repetitive paths, and avoids obstacles, while satisfying the kinematic constraints of the robotic vacuum cleaner. Simultaneously, the cleaning priority index is dynamically updated, and the path order of unfinished tasks is adjusted in real time to ensure that high-demand areas are prioritized and thoroughly cleaned.

[0096] Therefore, it can improve the cleaning effectiveness of robot vacuum cleaners while reducing ineffective energy consumption, and has good environmental adaptability and task flexibility.

[0097] In one specific embodiment, for grid cells that have not received new local observation data for a long time, an exponential decay operation is performed based on the time interval since their last update, allowing old pollution beliefs to weaken naturally over time and avoiding misjudgments due to outdated information. Furthermore, the number of actual pollution events occurring in each grid cell is continuously counted, and combined with the cumulative observation duration, the average pollution incidence rate per unit time is dynamically calculated by incrementally updating the event count and total time. This Poisson update mechanism can learn and remember the periodic or sudden characteristics of high-frequency pollution areas, thereby strengthening the recognition of high-risk areas while weakening old beliefs, ultimately achieving intelligent evolution and adaptive adjustment of the potential field in the spatiotemporal dimension.

[0098] The technical solution of this disclosure will be further explained below with a specific implementation example.

[0099] Based on the environmental sensors or recognition modules of the robotic vacuum cleaner, multi-source attribute observation data (i.e., local observation data) are obtained, and an initial potential field matrix (i.e., environmental attribute potential field matrix) is constructed, which includes environmental attribute dimensions such as dirt, stains, odors, and materials.

[0100] Preferably, the multi-source attribute observation data is preprocessed by a feature extraction and confidence calculation module to obtain a normalized environmental attribute observation vector: in, Represents the environmental attribute observation vector. This indicates the probability of dust being present. This indicates the probability of a stain being present. This indicates the probability of an unpleasant odor being present. This represents the probability of risk associated with ground material, with each dimension representing the probability of observation of different environmental attributes.

[0101] A Bayesian / Poisson hierarchical recursive fusion model is employed to recursively fuse and update multi-source observation data to obtain the basic potential field (i.e., the basic environmental attribute potential field) for multi-attribute fusion. Specifically, a Beta-Binomial model is used for binary environmental attributes (e.g., ground dirt level), a Gamma-Poisson model is used for counted environmental attributes (e.g., ground stain distribution and / or odor concentration), and a Dirichlet-Multinomial model is used for multi-class environmental attributes (e.g., ground material). In each observation, the learning step size is adaptively adjusted based on the gain of the current observation information relative to historical observation information to improve the stability and response sensitivity of the potential field.

[0102] Specifically, for multi-class environmental attributes, the state of the environmental attribute is represented by maintaining a Dirichlet parameter; for binary environmental attributes, the state of the environmental attribute is represented by a log-odds or Beta parameter; and for counted environmental attributes, the state is represented by Gamma (…). The environmental attribute state is represented by ) and the observation is weighted and fused based on the observation confidence of the current local observation data.

[0103] Preferably, the binary environment attributes are updated using the following adaptive step size. This represents the environmental attribute state at time t. This represents the environmental attribute state at time t-1. This represents the logarithmic probability increment. Indicates the probability of observation. represents the adaptive gain factor, and c represents the observation confidence of the current grid cell. This represents the local information entropy.

[0104] Preferably, the counting environment attributes are updated using Gamma–Poisson, i.e. Used to estimate the local contamination reach rate. Represents a cumulative count variable. Indicates the increment coefficient. Indicates a change in duration. Indicates the time step.

[0105] Spatial diffusion operations are performed on the fundamental potential field, including: Based on multi-scale Gaussian filtering and local entropy weighted fusion, local smoothing and structure preservation are achieved. Multi-scale convolutional diffusion calculates local entropy at different scales and performs entropy-weighted Gaussian fusion to simultaneously preserve local peaks and wide-area trends. This is applied to the fundamental potential field. At several scales Gaussian convolution is performed on each of the above. , This represents the smoothed basic potential field. Represents the Gaussian kernel function, calculates the local entropy of each grid cell, and assigns weights accordingly. merge into , This represents the adaptive weights at the k-th scale. Indicated in scale The local entropy of (i,j) under k, This represents the k-th spatial scale parameter. Indicates a scale index variable. Indicates the attenuation coefficient. This represents the basic potential field after first-level fusion. This step preserves small but strong stain peaks while reflecting the neighborhood accumulation trend, reducing the loss or oversmoothing problems caused by single-scale fusion.

[0106] Based on the potential field gradient and observation confidence level, the diffusion direction and intensity are controlled. Confidence-gated anisotropic diffusion (CGAD) introduces a confidence-gated function. , Indicates the base trust offset. This represents the confidence level of the original observations. It is combined with the Perona–Malik diffusion function. , This represents the gradient magnitude of the fundamental potential field. This represents the sensitivity threshold parameter. It implements adaptive diffusion control based on confidence level and gradient magnitude. Its anisotropic diffusion PDE is as follows: in, Indicates anisotropic diffusion. This represents the gradient vector of the fundamental potential field. Confidence-gated function used for boundary protection Incorporate observation confidence into the diffusion coefficient. Discrete implementation of neighborhood difference updates. , This represents the second-order optimal potential field. This indicates an update to the gain coefficient. This represents the local neighborhood difference term. It enables strong smoothing and diffusion in the high confidence region, while preserving details in the low confidence or large gradient regions, preventing noise mispropagation, and retaining material / structural boundaries.

[0107] Furthermore, Graph Laplace Directional Diffusion (GLDD) is achieved by establishing a weighted adjacency matrix. , Represents edge weight, Represents the squared Euclidean distance between nodes. This represents the decay rate parameter. Let represent the unit direction vector from node i to node j. This represents the dominant motion direction vector of node i. Indicates the directional enhancement coefficient. This represents the material continuity factor. A Laplace function is then constructed. , Represents the graph Laplace matrix. Depiction matrix, Let represent the adjacency matrix. Directional diffusion is approximated as... It spreads in the form of […]. This represents the potential field of the target environment after spatial diffusion. This represents the parameters of the diffusion event. It causes the potential field to preferentially diffuse along the actual direction of traffic or pedestrian flow and be suppressed by obstacles / material discontinuities, simulating the real-world spread of pollution.

[0108] Furthermore, after the robot vacuum performs its cleaning action, the probability of successful cleaning is calculated using a lightweight classification model based on cleaning feedback features (including changes in suction power, side brush vibration, image differences, and motor current), and the target environment attribute potential field is updated in a probability-weighted manner. Indicates the state of environmental attributes. This indicates the probability of a successful cleaning. This represents the cleaning baseline or a target value set according to the cleaning depth. This method avoids treating cleaning as a binary process, reducing erroneous downscaling of the potential field caused by misjudgments. Specifically, if the current observation confidence is low, a portion of the potential field strength is maintained to prevent misjudgment, thus forming a self-learning closed loop based on cleaning behavior feedback.

[0109] In one specific embodiment, a conditionalized residual kernel function is constructed based on the cleaning mode, floor material, and direction of travel of the robotic vacuum cleaner: The residual kernel function represents the kernel offset relative to the current raster cell. The relative intensity of pollutants that may remain at a given location due to incomplete cleaning; This represents the isotropic Gaussian attenuation term. Represents the offset vector. Indicates the dominant direction vector. This represents the directional modulation gain coefficient. It is used for neighborhood redistribution of undisturbed residues, reflecting the characteristics of pollution migration and residue diffusion.

[0110] Residual amount According to normalized nuclei The neighborhood assigned to the target raster cell: . This indicates the probability of a successful cleaning. Indicates the amount of pollutant residue. This indicates the environmental attribute state of the current raster cell at the previous moment. This indicates the environmental attribute state of the adjacent grid cells of the current grid cell. This step incorporates cleaning side effects (stain migration or diffusion) into the potential field update, improving the robustness of subsequent policies.

[0111] Simultaneously, a convolution operation is performed on the local potential field: This represents the potential field after convolution filtering. This represents the potential field after spatial diffusion processing. It enables the redistribution of the neighborhood and regeneration of the potential field for areas with remaining dirt. This mechanism simulates the diffusion and residue effects of stains in real-world scenarios, making subsequent tasks for the robotic vacuum cleaner more targeted.

[0112] To enable the potential field to adapt over a long period, a time decay and Poisson update mechanism is introduced. Exponential decay is applied to grid cells that have not received new observations for an extended period. Represents the current grid cell at time t. Environmental attribute status, Represents the grid cell at time t-1 Environmental attribute status, This represents the exponential decay factor.

[0113] In addition, a time decay and Poisson adaptive update mechanism is used to maintain the event frequency model: This indicates the estimated rate of pollution occurrence.

[0114] The long-term evolution and stable distribution of the potential field are achieved by performing spatiotemporal Laplace regularization or graphical Gaussian smoothing using the following formula: in, This represents the target environment potential field as the final output. The kernel matrix is ​​an adjacency matrix to ensure spatial continuity. Represents the variance of observation noise. This represents the identity matrix. After this stage, the potential field exhibits a decaying trend in time and stable diffusion in space, achieving self-organized evolution.

[0115] Specifically, the spatiotemporal evolution update includes an event-driven Poisson rate adjustment (EDPA) mechanism. When the frequency of observed events in a local region exceeds the expected value within a time window, the accumulation rate of the corresponding potential field is automatically increased and the revisit period is shortened. Gamma–Poisson parameters are maintained for each grid cell. In observation count The data is then updated. If the estimated rate of pollution occurrence exceeds the historical expected threshold in the short term, the potential field accumulation factor for that region is increased and the revisit interval is shortened; meanwhile, areas without observation for extended periods decay exponentially. This mechanism allows the potential field to reflect both long-term stable distributions and to trigger rapid responses to sudden, high-frequency pollution events.

[0116] Preferably, a cleaning priority index (CPI) is generated based on the final output target environmental attribute potential field: Indicates the cleaning priority index. Let represent the final potential value of the i-th type of environmental attribute. This represents the weight coefficient of the i-th type of environmental attribute. Gradient descent path planning is performed based on the cleaning priority index distribution, allowing the robot vacuum to prioritize cleaning high-priority areas, thus achieving differentiated strategy control. This mechanism enables the robot vacuum to possess semantic understanding and adaptive decision-making capabilities, achieving long-term, efficient cleaning in complex environments.

[0117] In one specific embodiment, the target environment attribute potential field is output from the final output. Generate hotspot clusters, i.e., connected component parts, and assign cleaning modes and paths. Extract connected component parts. where , This represents a set of highly polluted connected regions. This represents the k-th connected contaminated region. This indicates the pollution detection threshold.

[0118] Calculate the cleaning priority index (CPI) for each cluster: in, Indicates the stain intensity weight. This indicates a reward for the existence of connected regions. This represents the trend adjustment gain coefficient, where trend represents the potential field growth rate (historical difference), used to prioritize processing regions that are rising rapidly.

[0119] If the potential field of dirt within the cluster is high and the material allows, the cleaning mode is mopping and multiple coverings; if only the dirt is low-density, the cleaning mode is quick vacuuming. The planning algorithm can be A... The algorithm uses negative potential field values ​​as a heuristic (attracting high values) and employs a local coverage strategy.

[0120] Based on any of the above embodiments, this disclosure also provides an environment construction apparatus.

[0121] Figure 8 This is a schematic block diagram of the structure of an environment construction apparatus according to one embodiment of the present disclosure.

[0122] like Figure 8 As shown, the environment construction device includes: The data acquisition module 8002 acquires local observation data of different environmental attributes collected by the robot vacuum cleaner for the area to be cleaned. The environmental attributes include the ground material, the degree of dirt on the ground, the distribution of dirt on the ground and / or the concentration of odor. The data processing module 8004 preprocesses local observation data of different environmental attributes to determine the environmental attribute observation vectors of different environmental attributes. The preprocessing includes feature extraction, normalization and confidence calculation. The data mapping module 8006 maps the environmental attribute observation vector to the corresponding grid cell of the global grid map, and determines the environmental attribute potential field matrix of different environmental attributes. The global grid map is obtained by dividing the area to be cleaned into grids. The data fusion module 8008 recursively fuses the environmental attribute states of the grid cells in the corresponding environmental attribute potential field matrix based on the environmental attribute observation vectors of different environmental attributes to determine the basic environmental attribute potential field. The environmental attribute states are determined based on the historical environmental attribute observation vectors.

[0123] The data diffusion module 8010 performs spatial diffusion processing on the basic environmental attribute potential field to determine the target environmental attribute potential field for different environmental attributes.

[0124] The aforementioned environment construction device can be in the form of computer software, and each module of the aforementioned environment construction device can be implemented through computer software modules.

[0125] The specific implementation process of the functions and roles of each module in the above-mentioned environment construction device can be found in the implementation process of the corresponding steps in the above method, and will not be repeated here.

[0126] This disclosure also provides an electronic device 1000. Figure 9 A schematic diagram of the hardware implementation using the processing system is shown.

[0127] The hardware structure of electronic device 1000 can be implemented using a bus architecture. The bus architecture can include any number of interconnect buses and bridges, depending on the specific application and overall design constraints of the hardware. Bus 1100 connects various circuits including one or more processors 1200, memory 1300, and / or hardware modules. Bus 1100 can also connect various other circuits 1400 such as peripherals, voltage regulators, power management circuits, external antennas, etc. Bus 1100 can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Component (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, only one connection line is used in this figure, but this does not indicate that there is only one bus or one type of bus.

[0128] For ease of explanation, certain steps of the above method are described in relation to modules. It should be understood that the corresponding module performing one or more steps of the above method may be one or more hardware modules specifically configured to perform the corresponding step, or implemented by a processor configured to perform the corresponding step, or stored in a computer-readable medium for implementation by a processor, or implemented by some combination thereof.

[0129] According to an embodiment of this application, a robotic vacuum cleaner is also provided. This robotic vacuum cleaner includes the aforementioned electronic device.

[0130] This disclosure also provides a readable storage medium storing a computer program that, when executed by a processor, is used to implement the methods described above. A "readable storage medium" can be any means capable of containing, storing, communicating, propagating, or transmitting a program for use by or in conjunction with an instruction execution system, apparatus, or device. More specific examples of a readable storage medium include: an electrical connection with one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and programmable read-only memory (EPROM or flash memory), fiber optic devices, and portable read-only memory (CDROM), etc.

[0131] This disclosure also provides a computer program product, the methods of which can be implemented wholly or partially through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented wholly or partially as a computer program product. The computer program product includes one or more computer programs or instructions. When the computer program or instructions are loaded and executed, all or part of the processes or functions of this disclosure are performed.

[0132] Computer programs or instructions can be stored in a readable storage medium or transferred from one readable storage medium to another. For example, the computer program or instructions can be transferred from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means. The readable storage medium can be any available medium capable of access, or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium, such as a floppy disk, hard disk, or magnetic tape; an optical medium, such as a digital video optical disc; or a semiconductor medium, such as a solid-state drive. The computer-readable storage medium can be a volatile or non-volatile storage medium, or it can include both volatile and non-volatile types of storage media.

[0133] Those skilled in the art will understand that embodiments of this disclosure can be provided as methods, systems, or computer program products. Therefore, this disclosure can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this disclosure can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0134] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to this disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0135] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0136] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0137] In the description of this specification, the references to terms such as "one embodiment / mode," "some embodiments / modes," "example," "specific example," or "some examples," etc., refer to specific features, structures, or characteristics described in connection with that embodiment / mode or example, which are included in at least one embodiment / mode or example of this disclosure. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment / mode or example. Moreover, the specific features, structures, or characteristics described may be combined in any suitable manner in one or more embodiments / modes or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments / modes or examples described in this specification, as well as the features of different embodiments / modes or examples.

[0138] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this disclosure, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0139] Those skilled in the art should understand that the above embodiments are merely for illustrating the present disclosure and are not intended to limit the scope of the disclosure. Those skilled in the art can make other changes or modifications based on the above disclosure, and these changes or modifications still fall within the scope of the present disclosure.

Claims

1. An environmental modeling method, characterized in that, include: For the area to be cleaned, local observation data of different environmental attributes collected by the sweeping robot are obtained. The environmental attributes include ground material, degree of ground dirt, distribution of ground stains and / or odor concentration. Local observation data with different environmental attributes are preprocessed to determine environmental attribute observation vectors for different environmental attributes. The preprocessing includes feature extraction, normalization, and confidence calculation. The environmental attribute observation vectors are mapped to the corresponding grid cells of the global grid map to determine the environmental attribute potential field matrix for different environmental attributes. The global grid map is obtained by dividing the area to be cleaned into grids. Based on environmental attribute observation vectors with different environmental attributes, the environmental attribute states of grid cells in the corresponding environmental attribute potential field matrix are recursively fused to determine the basic environmental attribute potential field. The environmental attribute states are determined based on historical environmental attribute observation vectors. Spatial diffusion processing is performed on the basic environmental attribute potential fields to determine the target environmental attribute potential fields for different environmental attributes.

2. The environmental modeling method as described in claim 1, characterized in that, Based on environmental attribute observation vectors with different environmental attributes, the environmental attribute states of the corresponding grid cells in the environmental attribute potential field matrix are recursively fused, including: When the environmental attribute is the degree of ground dirtiness, at each time step, the observation probability in the corresponding environmental attribute observation vector is converted into a log-odds increment. The logarithmic probability increment is recursively fused with the environmental attribute state of the grid cell in the environmental attribute potential field matrix to determine the basic environmental attribute potential field.

3. The environmental modeling method as described in claim 1, characterized in that, Spatial diffusion processing is performed on the basic environmental attribute potential fields to determine the target environmental attribute potential fields for different environmental attributes, including: Based on the local observation data collected at the location of each grid cell in the basic environmental attribute potential field, the observation confidence of each grid cell is determined. The spatial diffusion intensity is determined based on the gate function constructed from the observation confidence level. The gradient magnitude of the basic environmental property potential field is calculated to determine the spatial diffusion boundary; Based on the spatial diffusion intensity and the spatial diffusion boundary, the basic environmental attribute potential field is subjected to spatial diffusion processing to determine the target environmental attribute potential field for different environmental attributes.

4. The environmental modeling method as described in claim 1, characterized in that, Spatial diffusion processing is performed on the basic environmental attribute potential fields to determine the target environmental attribute potential fields for different environmental attributes, including: Using grid cells as nodes, adjacent grid cells are connected by edges. The edge weights are determined by the physical distance between grid cells, the direction of pedestrian flow, and the continuity of ground material, thus determining the adjacency graph of the basic environmental attribute potential field. Spatial diffusion processing is performed based on the adjacency graph to determine the potential field of target environmental attributes for different environmental attributes.

5. The environmental modeling method as described in claim 1, characterized in that, After determining the potential field of the target environment attribute for different environmental attributes, the following is included: Real-time data collection of motor current changes, suction fluctuations, and visual frame differences of the robotic vacuum cleaner; The motor current change, suction fluctuation and visual frame difference are weighted and fused using a nonlinear regression model to determine the probability of successful cleaning. Based on the cleaning success probability, the environmental attribute state of the grid cells in the target environmental attribute potential field is updated.

6. The environmental modeling method as described in claim 5, characterized in that, After determining the potential field of the target environment attribute for different environmental attributes, the following is included: Based on the characteristics of the ground material, the movement direction of the sweeping robot, and the cleaning mode, a residual kernel function is constructed. For the target grid cell where the success rate of cleaning is less than the target probability, the amount of pollutant residue is calculated based on the success rate of cleaning and the environmental attribute status. The residual amount of pollutants is weighted and allocated with the residual kernel function to update the environmental attribute status of adjacent grid cells of the target grid cell.

7. The environmental modeling method as described in claim 1, characterized in that, After determining the potential field of the target environment attribute for different environmental attributes, the following is included: The environmental attribute states of each grid cell in the target environmental attribute potential field with different environmental attributes are weighted and calculated to determine the cleaning priority index of each grid cell in the target environmental attribute potential field. Based on the cleaning priority index, a path planning algorithm is used to plan the cleaning task of the sweeping robot.

8. An electronic device, characterized in that, include: The memory stores execution instructions; as well as A processor that executes the execution instructions stored in the memory, causing the processor to perform the environment modeling method according to any one of claims 1 to 7.

9. A robotic vacuum cleaner, characterized in that, The robotic vacuum cleaner includes the electronic device described in claim 8.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the environment modeling method according to any one of claims 1 to 7.

Images