System and method for performing a cleaning operation
Patent Information
- Application Number
- IN202541106891
- Authority / Receiving Office
- IN · IN
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2025-11-04
- Publication Date
- 2026-10-01
- Estimated Expiration
- 2045-11-04
AI Technical Summary
Conventional dishwashers require manual arrangement of utensils, consume excessive water, and lack adaptability to utensil geometry and surface contamination, leading to inefficient cleaning and resource wastage.
A system utilizing sensors and AI-driven robotic actuation to construct a 3D geometric model of objects, enabling adaptive cleaning with a scrubbing unit and water jet, and controlled hot air drying based on real-time feedback for efficient and geometry-aware cleaning and drying.
Achieves faster, adaptive, and resource-efficient cleaning and drying of utensils by automatically adjusting to their geometry and contamination levels, reducing operational time and ensuring thorough cleaning without damage.
Abstract
Description
TECHNICAL FIELD
[0001] The present disclosure relates to the field of cleaning technologies. Moreparticularly, the present disclosure relates to a system and method of performing acleaning operation.BACKGROUND
[0002] Conventional dishwashers rely on bulk water spraying mechanisms, whereutensils must be manually arranged within a rack before initiating a cleaning cycle.Existing dishwashers consume huge volumes of water mixed with detergent, that issprayed uniformly irrespective of utensil type or surface contamination. Thisgeneralized cleaning approach is inefficient, often requiring long operationaldurations ranging from 60 to 240 minutes to achieve satisfactory results. Moreover,the necessity for manual arrangement, loading, and unloading of utensils increasesuser effort and time. The process also fails to adapt to the geometry or material ofeach utensil, leading to inconsistent cleaning outcomes and resource wastage.
[0003] To address these limitations, the present disclosure provides a novelsystem and method that overcomes shortcomings of the prior art.OBJECTS OF THE PRESENT DISCLOSURE
[0004] It is a primary object of the present disclosure to provide a system thatachieves geometry-adaptive cleaning using sensor-guided robotic actuation.
[0005] It is another object of the present disclosure to provide a system thatautomates drying of objects through controlled hot air circulation and monitoring.SUMMARY
[0006] The present disclosure relates to the field of cleaning technologies. Moreparticularly, the present disclosure relates to a system and method of performing acleaning operation.
[0007] In an aspect, a system for performing a cleaning operation is disclosed.The system may include a processor and a memory coupled to the processor. Thememory may include processor-executable instructions, which, on execution, causethe processor to execute a sequence of tasks. The system may receive data selectedfrom any or a combination of a spatial measurement and one or more images of oneor more objects from one or more sensors. The system may further derive any or acombination of a depth map and a contour point cloud representation of the one ormore objects by applying data fusion and spatial correlation techniques to thespatial measurement and the one or more images of the one or more objects. Thesystem may construct a three-dimensional geometric model of the one or moreobjects using any or a combination of the depth map and the contour point cloudrepresentation of the one or more objects, the three-dimensional geometric modelrepresenting any or a combination of a contour, a depth, and a curvature of the oneor more objects. The system may create one or more motion trajectories for anactuator assembly, the one or more motion trajectories being determined based onthe constructed three-dimensional geometric model of the one or more objects byimplementing any or a combination of path-planning techniques, inversekinematics computations, and trajectory optimization models trained on objectgeometry datasets, and the actuator assembly being facilitated to position, orient,and actuate one or more cleaning elements to move around and within the one ormore objects based on the determined motion trajectories.BRIEF DESCRIPTION OF DRAWINGS
[0008] FIG. 1 illustrates an exemplary block diagram representation of a systemfor performing a cleaning operation, in accordance with an embodiment of thepresent disclosure.
[0009] FIGs. 2A-2C illustrate exemplary representations of the system, inaccordance with an embodiment of the present disclosure.
[0010] FIG. 3 illustrates an exemplary flowchart representation of a method ofperforming a cleaning operation, in accordance with an embodiment of the presentdisclosure.DETAILED DESCRIPTION
[0011] In an aspect, a system for cleaning objects, including utensils is disclosed.The system may include one or more sensors for obtaining spatial measurementsand images of one or more objects placed within a cleaning chamber. The receivedsensor data may be processed to construct a three-dimensional geometric model ofeach object. Based on the constructed model, one or more motion trajectories maybe created for an actuator assembly mounted with a scrub and a water jet. Theactuator assembly may follow the generated trajectories to perform cleaningoperations by removing residues and contaminants from the surfaces of the objects.The system may regulate the supply and pressure of water during the cleaningprocess, while a hot air mechanism may be used to remove excess water and drythe surfaces. The system may operate autonomously using artificial intelligencetechniques for path optimization, surface recognition, and cleaning efficiencyenhancement, thereby ensuring a faster and more adaptive washing process.
[0012] FIG. 1 illustrates an exemplary block diagram representation of a systemfor performing a cleaning operation, in accordance with an embodiment of thepresent disclosure.
[0013] Illustrated in FIG. 1 is a block diagram representation of the system forperforming a cleaning operation 100 (hereafter referred to as system 100).
[0014] In an embodiment of the present disclosure, the system 100 may include aprocessing engine 102. The processing engine 102 may include a processor 102-2operatively coupled with a memory 102-4. The processor 102-2 may include oneor more microprocessors, Digital Signal Processors (DSPs), Application-SpecificIntegrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), SystemOn-Chip (SoC) architectures, or any combination thereof, to execute machinereadable instructions stored in the memory 102-4. The processor 102-2 may enableautonomous cleaning, drying, and motion control based on object geometryreconstructed from multi-sensor data. The memory 102-4 may include one or morenon-transitory, computer-readable storage media, including, but not limited to,Random-Access Memory (RAM), Read-Only Memory (ROM), flash memory,magnetic storage, optical storage, or any combination thereof.
[0015] In an embodiment of the present disclosure, the processor 102-2 may beoperatively coupled with one or more sensors 104. The sensors 104 may include adistance measuring sensor 104-2 and a camera sensor 104-4. The sensors 104 maycapture spatial and visual information of one or more objects placed within thecleaning chamber. The distance measuring sensor 104-2 may provide depth andcontour measurements, which may be used by the processor 102-2 to understandthe geometry of the objects. The camera sensor 104-4 may capture images that maybe processed to identify surfaces, edges, and regions requiring focused cleaning.The processor 102-2 may combine the outputs of the sensors 104 to construct athree-dimensional model of the objects, which may guide an actuator assembly 106during cleaning. The three-dimensional model of the object may be constructed byusing the fused data from the sensors 104. The processor 102-2 may convert depthmeasurements and image features into a contour point cloud representing thesurface geometry of the object. Surface reconstruction algorithms including, but notlimited to, Poisson surface reconstruction, Delaunay triangulation, ball-pivoting,and volumetric grid-based methods may then be applied to connect the points intocontinuous meshes, capturing both external and internal contours. The processor102-2 may perform interpolation and smoothing to refine the model and reducenoise or missing data. Machine learning models trained on geometric datasets mayenhance the accuracy of shape representation. The system 100 may use this 3Dmodel to guide motion planning and adaptive cleaning by the actuator assembly106. CNN models may process images and depth data captured by the sensors 104to identify relevant features of one or more objects. The CNN may extract spatialhierarchies of features, including edges, contours, and surface textures, which maybe used to distinguish different regions of each object. The processor 102-2 mayuse the extracted features to generate a detailed representation of the object'ssurfaces, which may improve accuracy in three-dimensional reconstruction. Themodel may learn to associate depth variations with specific geometric patterns,enabling precise mapping of concave and convex areas. Feature maps generated bythe CNN may be fused with distance measurements from the distance measuringsensor 104-2 to refine surface localization. The system 100 may iteratively updatethe three-dimensional model using CNN predictions, reducing errors caused byocclusions or complex shapes. Implementing the CNN model may enhance theability of the system 100 to recognize surfaces accurately and guide adaptivecleaning trajectories. The sensors 104 may also include one or more environmentalsensing units including a temperature sensor, a humidity sensor, and a surfacemoisture sensor. The temperature sensor may detect the ambient and internal airtemperature during drying sessions. The humidity sensor may measure a relativehumidity of the objects to assess drying progress. The surface moisture sensor maydetermine a residual water level on object surfaces, enabling the processor 102-2 toadjust drying parameters dynamically. This may be required to ensure that thedrying process achieves complete and uniform moisture removal from objectsurfaces. The surface moisture sensor may provide real-time feedback on residualwater levels, allowing the processor 102-2 to adjust the temperature, airflow, andduration of the hot air drying unit 108. Without this feedback, some areas of theobject may remain wet, leading to incomplete drying or potential bacterial growth.The system 100 may use this dynamic adjustment to optimize energy usage byavoiding unnecessary heating or extended drying cycles. This approach may alsoprevent overheating or damage to delicate objects while maintaining efficient andeffective drying performance.
[0016] In an embodiment of the present disclosure, the processor 102-2 may beoperatively coupled with the actuator assembly 106. The actuator assembly 106may include a scrubbing unit 106-2 and a water jet nozzle 106-4. The actuatorassembly 106 may perform mechanical cleaning actions based on motiontrajectories generated by the processor 102-2. The scrubbing unit 106-2 may applycontrolled mechanical force on the surface of an object to remove stains andresidues, while the water jet nozzle 106-4 may direct pressurized water to dislodgedebris and rinse the surface. The processor 102-2 may use real-time feedback fromthe sensors 104 to adjust the movement, pressure, and angle of operation of theactuator assembly 106 for optimal cleaning performance. The actuator assembly106 may work in coordination with the processor 102-2 to adapt its operation basedon the material and geometry of each object. The system 100 may implementReinforcement Learning (RL) models trained on cleaning performance datasets tooptimize the actuator trajectories and cleaning intensity in response to varyingsurface conditions detected by the sensors 104. RL models may be a class ofmachine learning models that learn optimal actions through trial-and-errorinteractions with an environment to maximize a defined reward function. In thesystem 100, RL models may use feedback from the sensors 104 to evaluate theeffectiveness of each cleaning action performed by the actuator assembly 106. Theprocessor 102-2 may assign higher rewards to actions that result in cleaner surfacesor efficient water usage, guiding the model to learn preferred behavior over time.The RL models may adjust actuator parameters, including movement speed,scrubbing force, and water jet pressure, based on real-time sensor feedback. TheRL models may continuously refine its policy to adapt to different utensil shapes,materials, and contamination levels. Data from multiple cleaning cycles may beused to improve prediction accuracy and decision-making efficiency. The RLmodels may enable the system 100 to dynamically optimize actuator trajectoriesand cleaning intensity for superior performance. In an exemplary embodiment, theactuator assembly 106 may be fitted with one or more integrated sensors 106-6 thatmay assist in determining the amount of pressure applied to the surface during thecleaning operation. The integrated sensors 106-6 may continuously monitor contactforce to ensure optimal cleaning efficiency without damaging the object. Theprocessor 102-2 may dynamically adjust actuator movement or applied pressurebased on real-time feedback from the sensors 104. In case any of the integratedsensors 106-6 transmits misleading or inconsistent data, the actuator assembly 106may adapt its operation by recalibrating the applied force using redundant sensorinputs or predictive algorithms, thereby maintaining consistent and safe cleaningperformance. Further, the system 100 may utilize hot water in addition to regularwater through the actuator assembly 106 to enhance the removal of grease andhardened residues from the utensil surface. In certain cleaning modes, the system100 may also employ steam air to sanitize utensils by exposing them to hightemperature vapor that eliminates microbial contaminants. The processor 102-2may regulate the temperature and duration of hot water or steam application basedon data received from the sensors 104, ensuring effective cleaning and disinfectionwithout damaging sensitive materials.
[0017] In an embodiment of the present disclosure, the processor 102-2 may beoperatively coupled with a hot air drying unit 108. The hot air drying unit 108 mayfacilitate the removal of residual moisture from the surfaces of the cleaned objectsafter completion of the washing process. The processor 102-2 may control thetemperature, pressure, and duration of hot air flow based on data received from thesensors 104 that detect surface wetness and ambient humidity. The hot air dryingunit 108 may direct air through adjustable vents to ensure uniform drying of theobjects with varying shapes and orientations. The actuator assembly 106 may assistin repositioning or orienting the objects to improve airflow exposure wherenecessary. The processor 102-2 may adjust the temperature, airflow, and durationof the hot air drying unit 108, which may include a heating element 108-2 and anair outlet nozzle 108-4 that releases a controlled stream of hot air to evaporateresidual moisture after pressurized air removes excess water, depending on the typeof utensil such as plastic, steel, or others. The system 100 may implement regressionensemble models trained on temperature-humidity correlation datasets to optimizedrying parameters dynamically, thereby minimizing energy consumption whileensuring complete moisture removal. Regression ensemble models may combinepredictions from multiple regression algorithms to improve the accuracy androbustness of drying parameter estimation in the system 100. The models mayanalyze input data received from the sensors 104, including temperature, humidity,and surface moisture readings, to determine optimal drying conditions. Theprocessor 102-2 may use an output of the regression ensemble models to adjust airtemperature, flow rate, and drying duration in real time. Each base model withinthe ensemble, including but not limited to linear regression, decision trees, andgradient boosting regressors, may contribute weighted predictions based on its pastperformance. The regression ensemble models may continuously learn from dryingcycles to refine the relationship between environmental parameters and dryingefficiency. The system 100 may apply feedback-driven recalibration to ensureconsistent results across objects including utensils of different materials andgeometries.
[0018] In an embodiment of the present disclosure, the processor 102-2 may beoperatively coupled with a power control unit 110. The power control unit 110 maymanage the distribution and regulation of electrical power among various functionalcomponents. The processor 102-2 may communicate with the power control unit110 to allocate appropriate power levels to the sensors 104, the actuator assembly106, and the hot air drying unit 108 based on operational requirements. The powercontrol unit 110 may monitor voltage and current levels to prevent overloads orinefficiencies during simultaneous operation of multiple subsystems. The powercontrol unit 110 may also enable adaptive power modulation to optimize energyusage during low-demand or standby conditions. The system 100 may employintelligent power management techniques to ensure stable, efficient, and safeoperation throughout the cleaning and drying cycles.
[0019] In an embodiment of the present disclosure, the processor 102-2 may beoperatively coupled with a communication interface 112. The communicationinterface 112 may enable data exchange between internal components and externalcontrol interfaces. The processor 102-2 may use the communication interface 112to transmit control signals and receive feedback from the sensors 104, the actuatorassembly 106, the hot air drying unit 108, and the power control unit 110. Thecommunication interface 112 may support wired or wireless communicationprotocols for real-time coordination of operations and remote monitoring. Thesystem 100 may employ secure data transmission techniques to maintain reliability,synchronization, and operational integrity across all connected components. In anexemplary and non-limiting embodiment, the communication interface 112 mayenable the system 100 to receive user inputs such as cleaning mode selection, objecttype, or drying preference through a connected control panel or mobile application.The communication interface 112 may facilitate bidirectional communicationbetween the user and the processor 102-2, allowing users to monitor system statusand modify operational parameters in real time. Being IoT-enabled, thecommunication interface 112 may connect the system 100 to a cloud network forremote control, performance analytics, and predictive maintenance. Thecommunication interface 112 may also transmit operational logs and sensor data forstorage, analysis, or integration with smart home or industrial automation platforms.
[0020] In an embodiment of the present disclosure, the system 100 may receivedata from the sensors 104. The sensors 104 may be positioned within a cleaningchamber, integrated with the system 100, to capture the characteristics of one ormore objects placed in the cleaning chamber for a cleaning operation. The objectsmay include utensils, dishes, bowls, plates, cups, cutlery, cooking vessels, and otherwashable kitchen items of varying shapes, sizes, and materials. The distancemeasuring sensor 104-2 may provide spatial measurements in the form of depth orcontour data corresponding to the utensils. The camera sensor 104-4 may captureimages representing visual and structural features of the objects. The collectedspatial and image data may be transmitted to the processor 102-2 for analysis. Thesystem 100 may perform synchronized data acquisition to ensure temporalalignment between the spatial measurements and the captured images for accuratethree-dimensional analysis. The system 100 may perform synchronized dataacquisition by coordinating the operation of the sensors 104 to capture spatialmeasurements and images at closely matched time intervals. The processor 102-2may trigger sensor data capture using precise timing signals to ensure that depthreadings and visual images correspond to the same moment in time. The sensors104 may transmit their outputs to the processor 102-2 with minimal latency,allowing real-time alignment of multi-modal data. The processor 102-2 may applytemporal interpolation or buffering techniques to correct minor delays and maintainsynchronization. This process may ensure that the constructed three-dimensionalgeometric model accurately represents the object's shape and surface features.
[0021] In an embodiment of the present disclosure, the system 100 may employdata fusion and spatial correlation techniques to integrate spatial measurements andimages of the one or more objects to obtain a comprehensive spatial understanding.The sensors 104 may capture multi-modal data including, but not limited to, depth,texture, and geometric variations across the surface of the objects. The processor102-2 may perform spatial alignment of these heterogeneous datasets to ensurecorrespondence between image pixels and spatial coordinates. The fused data maybe processed using disparity estimation algorithms that may analyze parallaxdifferences across multiple image viewpoints. The processor 102-2 may then applyspatial correlation analysis to identify consistent spatial patterns and depthdiscontinuities in the measurement data. Using these derived correlations, thesystem 100 may generate a dense depth map representing distance gradients acrossthe object surfaces. Finally, the depth map may be transformed into a contour pointcloud representation that may capture the three-dimensional geometry andboundary features of the objects.
[0022] In an embodiment of the present disclosure, the system 100 may constructthe three-dimensional geometric model of the objects using the contour point cloudrepresentation of the one or more objects. The three-dimensional geometric modelmay represent any or a combination of a contour, a depth, and a curvature of theobjects. The contour point cloud representation of the objects may include a set ofspatially distributed points representing surface boundaries and geometric featuresof the objects captured by the sensors 104. The system 100 may analyze the pointcloud representation to identify surface continuity, edges, and curvature variationsacross the objects. Each point in the cloud may correspond to a specific coordinatein three-dimensional space, providing detailed information about shape andorientation. The processor 102-2 may apply surface reconstruction algorithms,including mesh generation or volumetric modeling techniques, to connect the pointsand create continuous surfaces. The system 100 may refine the model byinterpolating gaps or smoothing irregularities in the point cloud to enhancegeometric accuracy. The three-dimensional geometric model may represent bothexternal and internal contours, enabling precise mapping of concave and convexregions. The processor 102-2 may use the contour point cloud as a foundationaldataset to construct a high-fidelity three-dimensional model that may guidesubsequent cleaning operations.
[0023] In an embodiment of the present disclosure, the system 100 may create oneor more motion trajectories for the actuator assembly 106. Path-planning techniquesmay be used by the processor 102-2 to determine feasible routes for the actuatorassembly 106 to reach all surfaces of one or more objects without collisions. Pathplanning techniques may analyze the three-dimensional geometric model derivedfrom the sensors 104 to identify accessible regions and avoid obstacles. Inversekinematics computations may be applied to translate the desired end-effectorpositions along the planned paths into joint angles and actuator movements for theactuator assembly 106. Inverse kinematics computations may involve calculatingthe joint angles and actuator positions required for the actuator assembly 106 toreach a desired target position and orientation. The process may use mathematicalmodels to derive the relationship between the spatial coordinates of the actuatorassembly 106 and the movement of the scrubbing unit 106-2 and the water jetnozzle 106-4. The computations may rely on matrix transformations, trigonometricequations, or optimization algorithms to find feasible joint configurations thatachieve smooth and precise motion. The system 100 may employ real-time inversekinematics to adapt to dynamic environments or varying object positions detectedby sensors, ensuring accurate and collision-free operation. Trajectory optimizationmodels trained on object geometry datasets may predict the most efficient motionsequences that minimize cleaning time, energy consumption, or mechanical strain.These models may use historical data to refine actuator movements based on objectshapes, surface orientations, and cleaning requirements. The processor 102-2 maycombine outputs from path-planning, inverse kinematics, and trajectoryoptimization to generate smooth, collision-free motion trajectories. The actuatorassembly 106 may follow these trajectories to move the scrub and water jetprecisely over internal and external surfaces. Real-time adjustments may be madebased on feedback from the sensors 104 to account for variations in objectplacement or unexpected obstacles.
[0024] In an embodiment of the present disclosure, the actuator assembly 106 ofthe system 100 may be facilitated to position, orient, and actuate one or morecleaning elements including the scrubbing unit 106-2 and the water jet nozzle 106-4 by receiving motion commands from the processor 102-2 based on the determinedmotion trajectories. The processor 102-2 may translate the trajectories into precisejoint movements and end-effector orientations for the actuator assembly 106. Thesensors 104 may provide real-time feedback on object placement and surfaceconditions, which may be used to adjust actuator movements dynamically. Theactuator assembly 106 may enable scrubbing and jet spraying of water alonginternal and external surfaces of the objects to ensure comprehensive cleaning. Thesystem 100 may iteratively refine actuator operations using sensor feedback tomaintain accuracy and efficiency throughout the cleaning process.
[0025] In an embodiment of the present disclosure, the system 100 may enablesteam sanitization of one or more objects by activating the hot air drying unit 108to release controlled bursts of high-temperature steam. The sensors 104 may detectthe type, material, and contamination level of the objects, enabling the processor102-2 to determine the optimal steam temperature and duration. The hot air dryingunit 108 may include an internal steam generation mechanism that converts storedwater into pressurized steam, which may be directed toward the surfaces of theobjects. The high-temperature steam may penetrate crevices and irregular surfacesto eliminate bacteria, grease, and residual contaminants. The actuator assembly 106may position the objects or rotate them to ensure uniform steam exposure. Theprocessor 102-2 may monitor the process in real time to prevent overheating ormaterial degradation. Once sanitization is complete, the system 100 may initiate adrying cycle to remove remaining moisture.
[0026] In an embodiment of the present disclosure, the system 100 may monitora motion status of the actuator assembly 106 using feedback from the sensors 104.Upon detecting cessation of movement of the actuator assembly 106, the processor102-2 may generate control signals to actuate the pressurized hot air drying unit108. The processor 102-2 may detect cessation of movement of the actuatorassembly 106 by continuously monitoring feedback from the sensors 104, includingposition encoders, joint angle sensors, and velocity measurements. The actuatorassembly 106 may provide real-time data indicating its current location, orientation,and motion speed, which the processor 102-2 may analyze to determine whetherthe scrub and water jet nozzle have completed the planned motion trajectories. Theprocessor 102-2 may apply threshold checks on velocity or positional change, andwhen the detected motion falls below a minimal movement threshold for a specifiedduration, it may infer that the actuator assembly 106 has stopped. The actuatorassembly 106 may stop moving on the object after completing all cleaning paths,covering the entire surface area, and achieving the desired cleaning coveragedetermined by feedback from the sensors 104. The system 100 may also considersafety constraints, object geometry, and detected anomalies before confirmingmotion cessation, ensuring the actuator assembly 106 halts precisely at the end ofits cleaning task. This detection may trigger subsequent operations, such asactivation of the hot air drying unit 108, to begin the drying phase efficiently. Theprocessor 102-2 may determine optimal airflow, temperature, and durationparameters based on data from the sensors 104 indicating residual surface moistureand ambient conditions. The hot air drying unit 108 may then direct pressurized airover the objects to remove excess water and facilitate drying. The system 100 mayuse iterative feedback from the sensors 104 to adjust the drying operationdynamically, ensuring efficient and uniform moisture removal.
[0027] In an embodiment of the present disclosure, the system 100 mayimplement monocular depth prediction networks to estimate depth informationfrom single images captured by the sensors 104. Monocular depth predictionnetworks may be deep learning models that estimate depth information from asingle two-dimensional image. These networks may learn spatial and contextualrelationships between pixels to infer the relative distance of objects within a scene.A monocular depth prediction model may be trained on large datasets containingpaired RGB images and corresponding depth maps to recognize geometric patternsand surface variations. The system 100 may employ monocular depth predictionnetworks to generate depth estimations when only camera-based input from thesensors 104 is available. This approach may enable accurate three-dimensionalunderstanding without requiring stereo or LiDAR sensors. Monocular depthprediction networks may analyze visual cues, including texture gradients, shading,and perspective distortions, to predict relative distances of surfaces of the objectsto be cleaned. The processor 102-2 may generate a dense depth map of the objectsbased on network predictions. Point cloud registration techniques may then beapplied to align and merge depth data obtained from multiple viewpoints or sensormodalities. The processor 102-2 may use point cloud registration techniques tocorrect misalignments, reduce noise, and produce a consistent spatial representationof the object surfaces. The combined use of depth prediction and point cloudregistration may allow the system 100 to construct accurate three-dimensionalsurface maps, capturing both external and internal contours, and enabling precisemodeling of object geometry to guide adaptive cleaning operations.
[0028] In an embodiment of the present disclosure, Deep Q-Learning and policygradient models may be implemented by the system 100 to optimize cleaning pathsof the actuator assembly 106 based on learned policies. Deep Q-Learning and policygradient models may be reinforcement learning models that enable intelligentdecision-making based on reward feedback from the environment. Deep QLearning may use a neural network to approximate a Q-function that estimates theexpected cumulative reward for each possible action in a given state. Policygradient models may directly optimize a parameterized policy that determineswhich actions to take by maximizing an expected reward through gradient-basedlearning. The system 100 may implement Deep Q-Learning and policy gradientmodels to learn adaptive cleaning strategies that balance efficiency, coverage, andenergy use. Both models may continuously improve through iterative training andreal-time feedback. The models may receive state inputs from the sensors 104,including, but not limited to, object geometry, surface conditions, and prior cleaningoutcomes. Deep Q-Learning may estimate the expected cumulative reward forpossible actions, where the reward function may represent cleaning efficiency andsurface coverage. Policy gradient models may directly learn a probabilitydistribution over actions to maximize the expected reward during each cleaningcycle. The processor 102-2 may select actuator movements that achieve the highestpredicted rewards, dynamically adjusting the scrub and water jet trajectories. Themodels may continuously update their parameters using feedback from the sensors104 to improve decision-making for diverse object shapes and contamination levels.These reinforcement learning techniques may enable the system 100 to performadaptive, efficient, and geometry-aware cleaning of objects with optimizedcoverage and resource utilization.
[0029] In an embodiment of the present disclosure, the system 100 mayimplement Rapidly-exploring Random Tree (RRT) models to explore possiblemotion paths for the actuator assembly 106 in a high-dimensional space. RRTmodels may be path-planning models that efficiently search high-dimensionalspaces for feasible motion trajectories. RRT models may operate by incrementallybuilding a tree of nodes that randomly explore the environment while ensuringcollision-free paths between points. The system 100 may implement RRT modelsto determine optimal routes for the actuator assembly 106 to move around or withincomplex object geometries. RRT models may be particularly useful in dynamic orpartially constrained environments where deterministic planning is computationallyintensive. This technique may enable smooth, efficient, and adaptive movementplanning during cleaning operations. RRT models may generate a tree of randomsamples that may be connected to form feasible paths while avoiding collisions withobjects or cleaning chamber boundaries. RRT models may also be used toefficiently explore high-dimensional spaces for feasible motion trajectories. RRTmodels may operate by incrementally building a tree of randomly sampled pointsconnected through collision-free paths. RRT models may be particularly useful forrobotic motion planning where the environment is complex or partially constrained.RRT models may quickly find a valid path from a start point to a goal whileavoiding obstacles detected by the sensors 104. Dynamic obstacle avoidancetechniques may be applied to update the planned paths in real time based onmovements or changes detected by the sensors 104. Graph Neural Network (GNN)models trained on spatial constraint datasets may enhance the ability of the system100 to predict safe connections between nodes in the RRT and identify potentialcollision points. The processor 102-2 may use the GNN outputs to prioritize pathsthat maximize clearance and minimize travel time. The actuator assembly 106 mayfollow the optimized, collision-free trajectory while performing cleaning operationson the objects. These techniques may enable the system 100 to perform adaptive,safe, and efficient motion planning in complex and dynamic environments.
[0030] In an embodiment of the present disclosure, the system 100 may performgrime detection and surface cleanliness estimation of the objects by implementingsemantic segmentation models and U-Net architectures trained on annotated surfacecontamination datasets. Semantic segmentation models and U-Net architecturestrained on annotated surface contamination datasets may be deep learning modelsthat classify each pixel of an image based on contamination levels. These modelsmay learn to distinguish between clean and soiled regions by analyzing visual cues,including, but not limited to, color gradients, texture variations, and reflectivitypatterns. The U-Net architecture may employ an encoder-decoder structure thatextracts spatial features and reconstructs detailed segmentation maps for accuratecontamination localization. The system 100 may use these models to detect grime,stains, or residue on utensils during cleaning. This pixel-level analysis may enableprecise surface cleanliness estimation and adaptive control of the actuator assembly106. These models may work by classifying each pixel of an image captured by thesensors 104 into specific categories representing clean or contaminated regions. Thesystem 100 may employ U-Net architectures with encoder-decoder structures thatmay extract spatial and contextual features from utensil images to identify regionscontaining grime, stains, or residue. The processor 102-2 may process thesegmented output to generate a contamination probability map highlighting surfaceareas requiring further cleaning. The segmentation models may learn intricatevisual cues to differentiate between clean and soiled surfaces. The sensors 104 maycontinuously provide real-time visual feedback to refine detection accuracy. Theprocessor 102-2 may fuse multi-sensor inputs to enhance the reliability of thecleanliness estimation process. Based on this analysis, the actuator assembly 106may dynamically adjust cleaning trajectories and scrubbing intensity.
[0031] In an embodiment of the present disclosure, regression ensemble modelsand predictive analytics techniques trained on temperature-humidity correlationdatasets may work by estimating the most efficient drying parameters, including,but not limited to, air temperature, airflow velocity, nozzle orientation, dryingduration, and heat distribution uniformity, to achieve uniform moisture removal.The system 100 may employ ensemble learning methods that may combine multipleregression algorithms, including, but not limited to, random forests, gradientboosting, and Bayesian regression, to predict optimal air temperature, flow rate, andduration. The processor 102-2 may analyze inputs received from the sensors 104measuring real-time temperature, humidity, and surface moisture to dynamicallyadjust the operational state of the hot air drying unit 108. Predictive analyticstechniques may identify patterns between environmental factors and dryingperformance to anticipate the drying outcome before completion. The models maycontinuously learn from historical drying cycles to improve future performancethrough feedback optimization. The system 100 may thereby minimize energyconsumption while preventing overheating or under-drying. The adaptive controlmay ensure consistent drying efficiency across diverse utensil geometries andmaterials.
[0032] In an embodiment of the present disclosure, vision-based pose estimationmodels and Siamese neural networks trained on reference alignment datasets maywork by analyzing visual inputs from the sensors 104 to determine the spatialorientation and placement of the one or more objects. Vision-based pose estimationmodels and Siamese neural networks may be advanced computer vision models thatidentify and compare spatial orientations of objects in visual data. Vision-basedpose estimation models may analyze keypoints, contours, and geometric landmarksfrom images to determine a position and rotation of the objects in three-dimensionalspace. Siamese neural networks may consist of twin sub-networks that learn tomeasure similarity between two input images by comparing their featureembeddings. The system 100 may employ these models to verify the alignment andplacement of utensils before cleaning. This combined approach may ensureaccurate object positioning and prevent mechanical interference during operation.The system 100 may use pose estimation algorithms to extract key points andgeometric landmarks from object images, enabling precise calculation of translationand rotation parameters in three-dimensional space. The Siamese neural networksmay compare the detected object pose with reference models stored in a dataset toverify correct positioning before initiating the cleaning process. The processor 102-2 may process the similarity scores generated by the Siamese network to identifyany positional deviations. The sensors 104 may provide continuous visual feedbackto update alignment status in real time. The combined approach may enable high-accuracy verification even under varying lighting and reflective surface conditions.This verification process may ensure optimal cleaning trajectories and preventmechanical interference with the actuator assembly 106.
[0033] In an embodiment of the present disclosure, One-class SVM and LSTMbased temporal pattern recognition models trained on machine behavior andperformance datasets may work collaboratively to detect deviations from normaloperating conditions during the cleaning process. The system 100 may employ aone-class SVM to establish a decision boundary representing the normal behavioralpattern of the actuator assembly 106, based on features including, but not limitedto, vibration signatures, motor current, and torque feedback. The LSTM-basedtemporal models may analyze sequential data streams from the sensors 104 toidentify abnormal temporal fluctuations that deviate from expected operationaltrends. The processor 102-2 may fuse spatial and temporal anomaly indicators todetermine whether the detected behavior corresponds to mechanical obstruction,excessive load, or sensor malfunction. The models may continuously learn fromoperational feedback to refine anomaly thresholds and detection accuracy. Thesystem 100 may issue alerts or automatically adjust actuator parameters to preventdamage. This adaptive anomaly detection mechanism may ensure sustainedreliability and operational safety during cleaning.
[0034] FIGs. 2A-2C illustrate exemplary representations of the system, inaccordance with an embodiment of the present disclosure.
[0035] Illustrated in FIG. 2A is a representation 200A of the object 202 beingscanned by the sensors 104. The sensors 104 may scan the object 202 to be cleanedto acquire detailed spatial and visual information necessary for accurate cleaning ofthe object 202. The sensors 104 may include the distance measuring sensor 104-2and the camera sensor 104-4 that may capture depth and image data related to theobject 202 from multiple angles. The processor 102-2 may process the scanned datato construct a three-dimensional geometric model representing surface contours andinternal geometry of the object 202. The scanning may take place to enable preciseplanning of motion trajectories for the actuator assembly 106. This process mayensure targeted cleaning by adapting a motion path of the actuator assembly 106 toa unique shape of the object 202.
[0036] Illustrated in FIGs. 2B-2C are representations 200B-200C of animplementation of the actuator assembly 106. The actuator assembly 106 of thesystem 100 may work on the object 202 by executing targeted cleaning actionsbased on motion trajectories generated by the processor 102-2. The scrubbing unit106-2 may apply controlled mechanical force across the surface of the object 202to remove adhered grime, stains, and residue. The scrubbing unit 106-2 may operatethrough a combination of rotary and oscillatory cleaning mechanisms driven by anelectric or pneumatic actuator. The scrubbing unit 106-2 may include multi-texturedbrushes, microfiber pads, or abrasive discs that contact the utensil surface to removefood residues, oil, and stains. The scrubbing unit 106-2 may adjust its pressure androtation speed based on utensil type, contamination level, and feedback fromembedded load or friction sensors. Cleaning agents may be dispensed throughcontrolled nozzles integrated into the scrubbing assembly to enhance the removalefficiency. The scrubbing process may be coordinated by the processor 102-2 toensure uniform coverage and minimal material wear.
[0037] In an embodiment of the present disclosure, the water jet nozzle 106-4 maysimultaneously emit pressurized water streams to dislodge and rinse away loosenedparticles. The sensors 104 may continuously provide feedback on surface conditionand cleaning progress to optimize scrubbing intensity and jet pressure. The waterjet nozzle 106-4 may operate by directing high-pressure water streams toward theutensil surface to dislodge grease, stains, and residual particles. The water jet nozzle106-4 may include a precision orifice and an adjustable valve that regulate thevelocity, angle, and spread of the jet flow according to the cleaning phase. Thewater jet nozzle 106-4 may rotate or oscillate under actuator control to ensureuniform coverage across complex utensil geometries. The water jet nozzle 106-4may use feedback from proximity or contamination sensors to dynamically adjustpressure levels and spray patterns. The water jet nozzle 106-4 may also incorporatetemperature regulation to alternate between cold and heated water for optimizedcleaning efficiency and sterilization. This coordinated cleaning operation mayensure comprehensive coverage and efficient removal of contaminants from theobject 202.
[0038] FIG. 3 illustrates an exemplary flowchart representation of a method ofperforming a cleaning operation, in accordance with an embodiment of the presentdisclosure.
[0039] Illustrated in FIG. 3 is a representation of the method 300 of performingthe cleaning operation. The method 300 may begin with receiving 302, by theprocessor 102-2, data selected from any or a combination of a spatial measurementand one or more images of the objects 202 from the sensors 104. The method 300may proceed with deriving 304, by the processor 102-2, a contour point cloudrepresentation of the objects 202 by applying data fusion and spatial correlationtechniques to the spatial measurement and the one or more images of the objects202. The method 300 may proceed with constructing 306, by the processor 102-2,a three-dimensional geometric model of the objects 202 using the contour pointcloud representation of the objects 202. The three-dimensional geometric modelmay represent any or a combination of a contour, a depth, and a curvature of theobjects 202. The method 300 may proceed with determining 308, by the processor102-2, one or more motion trajectories for the actuator assembly 106. The one ormore motion trajectories may be determined based on the constructed threedimensional geometric model of the objects 202 by implementing any or acombination of path-planning techniques, inverse kinematics computations, andtrajectory optimization models trained on object geometry datasets. Further, theactuator assembly 106 may be facilitated to position, orient, and actuate thescrubbing unit 106-2 and the water jet nozzle 106-4 to move around and within theobjects 202 based on the determined motion trajectories.
[0040] In an embodiment of the present disclosure, the system 100 may initiateoperation when an object is placed within the cleaning chamber and detected by thesensors 104. The sensors 104 may include a combination of the distance measuringsensor 104-2 and the camera sensor 104-4 that may capture spatial measurementsand surface imagery from multiple viewpoints. The processor 102-2 may receivethis data and perform preprocessing operations to remove noise and distortions. Theprocessor 102-2 may implement data fusion and spatial correlation techniques tointegrate the inputs into a coherent three-dimensional representation. A contourpoint cloud representation of the object may be generated to define surfaceboundaries and depth variations. Using this data, the processor 102-2 may constructa detailed three-dimensional geometric model representing the physical shape andsize of the object. This geometric model may serve as a reference for trajectoryplanning and motion optimization. The scanning and modelling process may ensurethat every surface, curve, and cavity of the object is accurately represented.
[0041] In an embodiment of the present disclosure, once the three-dimensionalmodel is generated, the processor 102-2 may determine optimized motiontrajectories for the actuator assembly 106. The processor 102-2 may use pathplanning techniques, inverse kinematics computations, and trajectory optimizationmodels trained on object geometry datasets to define precise movement patterns.The actuator assembly 106 may include the scrubbing unit 106-2 and the water jetnozzle 106-4 designed for coordinated operation. The processor 102-2 may controlthe actuator assembly 106 to position, orient, and actuate the cleaning elementsalong the computed trajectories. The scrubbing unit 106-2 may apply mechanicalforce to dislodge grime, grease, and stains from the object's surface.Simultaneously, the water jet nozzle 106-4 may direct streams of pressurized waterto rinse and remove loosened contaminants. The sensors 104 may providecontinuous feedback regarding surface cleanliness, pressure distribution, andcleaning coverage. The processor 102-2 may dynamically adjust motion parametersof the actuator assembly 106 in response to this feedback. Reinforcement learningmodels trained on cleaning efficiency datasets may be employed to refine scrubbingintensity and water pressure for improved cleaning outcomes. This closed-loopcontrol process may enhance precision and adaptability during cleaning.
[0042] In an embodiment of the present disclosure, after the actuator assembly106 completes its cleaning cycle, the processor 102-2 may detect the cessation ofmotion and initiate the drying phase. The sensors 104 may measure temperature,humidity, and residual surface moisture to evaluate the object's dryingrequirements. The processor 102-2 may activate the hot air drying unit 108 to emitpressurized heated air uniformly across the cleaned object. Regression ensemblemodels trained on temperature-humidity correlation datasets may be used to predictoptimal air temperature and flow rate for efficient drying. The drying process mayremove residual moisture without damaging the object's surface. The processor102-2 may continue to monitor environmental parameters through the sensors 104to ensure consistent drying performance. Once the object reaches the desireddryness level, the system 100 may terminate operation and prepare for the nextcleaning cycle. This systematic process may ensure effective, adaptive, and energyefficient cleaning and drying of a wide range of utensils and dishware.
[0043] A use case of the system 100 is described herein. The system 100 maybegin operation when the object 202 that is a soiled utensil is placed within thecleaning chamber. The sensors 104 may scan the utensil to capture its geometry andsurface condition, allowing the processor 102-2 to generate an accurate threedimensional model. Based on this model, the processor 102-2 may control theactuator assembly 106 to direct the scrubbing unit 106-2 and the water jet nozzle106-4 along optimized paths to remove dirt and residue. The sensors 104 maymonitor cleanliness in real-time and provide feedback for adaptive cleaning. Afterwashing, the hot air drying unit 108 may emit pressurized heated air to ensurethorough drying of the utensil.ADVANTAGES OF THE PRESENT DISCLOSURE
[0044] The present disclosure provides a system that provides precise andefficient cleaning through adaptive robotic trajectory planning and control.
[0045] The present disclosure provides a system that minimizes water and energyconsumption by intelligent surface detection and cleaning optimization.
Claims
1. A system (100) for performing a cleaning operation, the system (100) comprising: a processor (102-2); and a memory (102-4) coupled to the processor (102-2), wherein the memory (102-4) comprises processor-executable instructions, which on execution, cause the processor (102-2) to: receive data selected from any or a combination of a spatial measurement and one or more images of one or more objects (202) from one or more sensors (104); derive a contour point cloud representation of the one or more objects (202) by applying data fusion and spatial correlation techniques on the received data; construct a three-dimensional geometric model of the one or more objects (202) using the contour point cloud representation of the one or more objects (202), the three-dimensional geometric model representing any or a combination of a contour, a depth, and a curvature of the one or more objects (202); and determine one or more motion trajectories for an actuator assembly (106), the one or more motion trajectories being determined based on the constructed three-dimensional geometric model of the one or more objects by implementing any or a combination of path-planning techniques, inverse kinematics computations, and trajectory optimization models trained on object geometry datasets, and the actuator assembly (106) being facilitated to position, orient, and actuate one or more cleaning elements to move around and within the one or more objects (202) based on the determined motion trajectories.
2. The system (100) as claimed in claim 1, wherein the processor (102-2) enables steam sanitization of the one or more objects (202) by activating a hot air drying unit (108) to release controlled bursts of high-temperature steam.
3. The system (100) as claimed in claim 1, wherein the processor (102-2) enables actuation of a hot air drying unit (108) upon detection of cessation of motion of the actuator assembly (106).
4. The system (100) as claimed in claim 1, wherein the processor (102-2) performs depth estimation and surface mapping of the one or more objects (202) by implementing monocular depth prediction networks and point cloud registration techniques.
5. The system (100) as claimed in claim 1, wherein the processor (102-2) executes adaptive cleaning path optimization for the actuator assembly (106) by implementing Deep Q-Learning and policy gradient models trained on reward functions representing cleaning efficiency and coverage.
6. The system (100) as claimed in claim 1, wherein the processor (102-2) performs collision-free motion planning for the actuator assembly (106) by implementing Rapidly-exploring Random Tree (RRT) models and dynamic obstacle avoidance techniques enhanced with Graph Neural Network (GNN) models trained on spatial constraint datasets.
7. The system (100) as claimed in claim 1, wherein the processor (102-2) performs grime detection and surface cleanliness estimation of the one or more objects (202) by implementing semantic segmentation models and U-Net architectures trained on annotated surface contamination datasets.
8. The system (100) as claimed in claim 1, wherein the processor (102-2) executes optimal drying cycle control by implementing regression ensemble models and predictive analytics techniques trained on temperature-humidity correlation datasets.
9. The system (100) as claimed in claim 1, wherein the processor (102-2) performs object positioning verification by implementing vision-based pose estimation models and Siamese neural networks trained on reference alignment datasets.
10. A method (300) for performing a cleaning operation, the method (300) comprising steps of: receiving (302), by a processor (102-2), data selected from any or a combination of a spatial measurement and one or more images of one or more objects (202) from one or more sensors (104); deriving (304), by the processor (102-2), a contour point cloud representation of the one or more objects (202) by applying data fusion and spatial correlation techniques to the spatial measurement and the one or more images of the one or more objects (202); constructing (306), by the processor (102-2), a three-dimensional geometric model of the one or more objects (202) using the contour point cloud representation of the one or more objects (202), the three-dimensional geometric model representing any or a combination of a contour, a depth, and a curvature of the one or more objects (202); and determine one or more motion trajectories for an actuator assembly (106), the one or more motion trajectories being determined based on the constructed three-dimensional geometric model of the one or more objects by implementing any or a combination of path-planning techniques, inverse kinematics computations, and trajectory optimization models trained on object geometry datasets, and the actuator assembly (106) being facilitated to position, orient, and actuate one or more cleaning elements to move around and within the one or more objects (202) based on the determined motion trajectories.
