Surface type detection and surface cleaning device using the same
By employing a microphone and machine learning model to detect surface types, the surface cleaning device efficiently adapts its cleaning behaviors to various surfaces, addressing the size and cost issues of traditional sensor configurations.
Patent Information
- Application Number
- US19/202686
- Authority / Receiving Office
- US · United States
- Patent Type
- Applications(United States)
- Current Assignee / Owner
- Filing Date
- 2025-05-08
- Publication Date
- 2026-11-12
AI Technical Summary
Existing surface cleaning devices face challenges in efficiently detecting and adapting to different surface types, such as carpets and hard surfaces, due to the inclusion of both emitters and detectors in surface type sensors, which increases size and cost.
The use of a microphone to detect sound generated by the operation of the surface cleaning device, combined with a machine learning model, to determine surface type without the need for an acoustic emitter, allowing for cost-effective and efficient surface type detection.
Enables accurate and adaptive cleaning behaviors based on surface type, reducing device size and cost while improving cleaning performance.
Smart Images

Figure US20260354641A1-D00000_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present disclosure is generally related to surface type detection in surface cleaning devices (e.g., vacuum cleaners) and more specifically to surface type detection using sound (or noise) generated through operation of a surface cleaning device.BACKGROUND INFORMATION
[0002] Surface cleaning devices are configured to clean one or more surfaces (e.g., floors) within a space to be cleaned. In some instances, the one or more surfaces may include a plurality of surface types, wherein at least one surface type is different from at least one other surface type. For example, a residential space, may include at least one carpeted surface type (e.g., high pile, low pile, and / or etc.) and at least one hard surface type (e.g., wood, tile, and / or etc.). In this example, a cleaning behavior of the surface cleaning device may be configured to be adjusted based on the surface type of the surface being cleaned (e.g., to improve cleaning performance). The cleaning behavior may be adjusted automatically or by a user using the surface cleaning device.
[0003] When adjusted automatically, the surface cleaning device may include a surface type sensor that is configured to detect the surface type of the surface being cleaned. The surface type sensor may be configured to generate an emission in a direction of the surface being cleaned and to detect a reflection of the generated emission. The detected reflection is used to determine the surface type. As such, the surface type sensor includes an emitter (e.g., an infra-red, IR, emitter or an ultrasonic emitter) and a detector (e.g., an IR detector or an ultrasonic detector). Inclusion of both an emitter and a detector within the surface type sensor may increase the size and / or cost of the surface type sensor.BRIEF DESCRIPTION OF THE DRAWINGS
[0004] These and other features and advantages will be better understood by reading the following detailed description, taken together with the drawings, wherein:
[0005] FIG. 1 is a schematic example of a surface cleaning device, consistent with embodiments of the present disclosure.
[0006] FIG. 2 is a schematic example of a robotic surface cleaning device, consistent with embodiments of the present disclosure.
[0007] FIG. 3 is a schematic side view of the robotic surface cleaning device of FIG. 2, consistent with embodiments of the present disclosure.
[0008] FIG. 4 is a schematic example of an upright surface cleaning device, consistent with embodiments of the present disclosure.
[0009] FIG. 5 is a schematic example of a surface type detection system, consistent with embodiments of the present disclosure.
[0010] FIG. 6 is a flow chart of an example training method, consistent with embodiments of the present disclosure.
[0011] FIG. 7 is a flow chart of an example operation method, consistent with embodiments of the present disclosure.
[0012] FIG. 8 is a perspective view of an example of a surface cleaning head for use with an upright cleaner, consistent with embodiments of the present disclosure.
[0013] FIG. 9 is a perspective view of another example of a surface cleaning head for use with an upright cleaner, consistent with embodiments of the present disclosure.
[0014] FIG. 10 is a bottom view of an example of a robotic surface cleaning device, consistent with embodiments of the present disclosure.DETAILED DESCRIPTION
[0015] The present disclosure is generally related to a surface cleaning device configured to detect a surface type of a surface being cleaned. The surface cleaning device includes a microphone configured to detect sound generated through operation of the surface cleaning device and a controller communicatively coupled to the microphone. An output of the microphone is provided to the controller, wherein the controller includes a machine learning model configured to determine a surface type based, at least in part, on the output of the microphone. The machine learning model may be generated through operation of a surface cleaning device on a surface having a known surface type for a test period while a microphone generates an output that is provided to a machine learning algorithm.
[0016] FIG. 1 is a schematic example of a surface cleaning device 100. The surface cleaning device 100 includes a main body 102, a dirty air inlet 104, a dust cup 106 fluidly coupled to the dirty air inlet 104, and a suction motor 108 fluidly coupled to the dust cup 106 and configured to cause air to flow into the dirty air inlet 104. Air flowing through the dirty air inlet 104 passes through the dust cup 106 and into the suction motor 108 to be exhausted into a surrounding environment. Air entering the dirty air inlet 104 may have debris entrained therein, wherein at least a portion of the entrained debris may fall out of entrainment within the dust cup 106 and be collected within the dust cup 106 for later disposal. As further shown, the surface cleaning device 100 may include an agitator 110 (e.g., a brush roll) configured to contact a surface to be cleaned 112 (e.g., a floor). Contact between the agitator 110 and the surface to be cleaned 112 may disturb at least a portion of any debris on the surface to be cleaned 112, allowing for easier entrainment of debris within air entering the dirty air inlet 104. Rotation of the agitator 110 may be caused by an agitator motor 114.
[0017] The surface cleaning device 100 may further include a controller 116. The controller 116 may be communicatively coupled to one or more components of the surface cleaning device 100. For example, the controller 116 may be communicatively coupled to the suction motor 108 and / or the agitator motor 114. In this example, the controller 116 may be configured to control rotation (e.g., a rotational speed) of one or more of the suction motor 108 and / or the agitator 110.
[0018] FIG. 2 shows a schematic example of a robotic surface cleaning device 200. The robotic surface cleaning device 200 is an example of the surface cleaning device 100 of FIG. 1. As shown, the robotic surface cleaning device 200 includes a main body 202, a dust cup 204 removably coupled to the main body 202, a dirty air inlet 206 fluidly coupled to the dust cup 204, a suction motor 208 fluidly coupled to the dirty air inlet 206 via the dust cup 204, and one or more driven wheels 210 configured to urge the robotic surface cleaning device 200 along a surface to be cleaned 212. The robotic surface cleaning device 200 further includes one or more sensors 214 and a controller 216 communicatively coupled to the one or more sensors 214 and the one or more driven wheels 210. In some instances, the robotic surface cleaning device 200 may further include one or more of an agitator 218 (e.g., a brush roll) configured to rotationally engage the surface to be cleaned 212 (e.g., to disturb debris on the surface to be cleaned 212), a side brush 220 configured to urge debris towards the dirty air inlet 206 and / or the agitator 218, and / or one or more air jets 222 configured to generate a jet of air in a direction extending outwardly from the robotic surface cleaning device 200.
[0019] The one or more sensors 214 may include one or more of a navigation sensor, an obstacle sensor, a surface type sensor, a drop off (or cliff) sensor, and / or any other type of sensor. Examples of the one or more sensors 214 may include one or more of cameras, light detection and ranging (LIDAR) sensors, time of flight sensors, infra-red (IR) proximity sensors, bump sensors (e.g., tactile switches), ultrasonic sensors, and / or any other type of sensor.
[0020] The controller 216 is configured to control operation of one or more of the one or more driven wheels 210, the suction motor 208, the agitator 218, the side brush 220, and / or the one or more air jets 222 based, at least in part, on outputs generated by the one or more sensors 214. For example, the one or more sensors 214 may be configured to detect one or more obstacles and the controller 216 may be configured to cause the one or more driven wheels 210 to maneuver the robotic surface cleaning device 200 to avoid the one or more obstacles. By way of further example, the one or more sensors 214 may be configured to detect a surface type of the surface to be cleaned 212 and the controller 216 may be configured to adjust a speed of the suction motor 208, the agitator 218, and / or the side brush 220 based, at least in part, on the detected surface type.
[0021] With additional reference to FIG. 3, in some instances, the robotic surface cleaning device 200 may include a ride height adjuster 224. The ride height adjuster 224 is configured to cause a ride height 300 (or separation distance) extending from a bottom surface 302 of the main body 202 to the surface to be cleaned 212 to increase or decrease. The controller 216 may cause the ride height 300 to increase (or decrease) in response to changes in surface type (e.g., as detected by the one or more sensors 214). Such a configuration may allow, for example, a suction force to be adjusted at the surface to be cleaned and / or prevent a wet component (e.g., a mop pad) from contacting a carpet. The ride height adjuster 224 may be configured to act on the one or more driven wheels 210 to urge the one or more driven wheels 210 to extend from or retract into the main body 202. For example, the ride height adjuster 224 may include a cam or lever to urge the one or more driven wheels 210 to extend from the main body 202.
[0022] When the one or more sensors 214 include a surface type sensor 226, the surface type sensor 226 can be positioned such that reflections from the surface to be cleaned 212 can be detected by the surface type sensor 226. For example, the surface type sensor 226 may be positioned proximate to the bottom surface 302 of the main body 202 and face the surface to be cleaned 212. By way of further example, the surface type sensor 226 may be disposed within the dirty air inlet 206 and proximate the agitator 218. In some instances, the surface type sensor 226 may be an acoustic sensor configured to detect sounds within an environment (e.g., sounds within the human audible spectrum). In these instances, sounds generated as a result of operation of the robotic surface cleaning device 200 (e.g., from operation of the suction motor 208, rotation of the agitator 218, rotation of the driven wheels 210, the rotation of the side brush 220, and / or any other component of the robotic surface cleaning device 200) may be reflected from the surface to be cleaned 212 and detected by the surface type sensor 226. Based, at least in part, on changes in output from the surface type sensor 226 resulting from the detected sounds, the controller 216 may be configured to determine a surface type. In this way, the surface type sensor 226 may be able to omit an acoustic emitter and only have an acoustic detector (e.g., a microphone).
[0023] In some instances, the controller 216 may be configured to generate a map using outputs from the one or more sensors 214 (e.g., one or more navigation sensors). The map may be representative of obstacles (e.g., walls and / or furniture) within an environment. The map may be stored and used for subsequent cleaning operations. In some instances, the map may include indications of surface type, wherein the indications of surface type are determined using the surface type sensor 226. As such, after generation of the map, the environment may be traversed using the map with or without use of the surface type sensor 226 and the behaviors of the robotic surface cleaning device 200 may be adjusted according to surface type based, at least in part, on the map and / or the output of the surface type sensor 226.
[0024] In some instances, the robotic surface cleaning device 200 may include a mopping assembly 228 removably coupled to the main body 202. The mopping assembly 228 may be configured for dry mopping (or dusting) and / or wet mopping.
[0025] FIG. 4 shows a schematic example of an upright surface cleaning device 400. The upright surface cleaning device 400 is an example of the surface cleaning device 100 of FIG. 1. As shown, the upright surface cleaning device 400 includes a surface cleaning head 402 and an upright section 404 extending from the surface cleaning head 402. The upright section 404 may be pivotally coupled to the surface cleaning head 402 about one or more axes (e.g., to allow side-to-side and / or forward / backward pivotal movement of the upright section 404). The surface cleaning head 402 includes a dirty air inlet 406 and the upright section 404 can include a dust cup 408 fluidly coupled to the dirty air inlet 406 and a suction motor 410 fluidly coupled to the dust cup 408. Operation of the suction motor 410 causes air to flow into the dirty air inlet 406 and through the dust cup 408. The surface cleaning head 402 can further include an agitator 412 (e.g., a brush roll) configured to agitate a surface to be cleaned 414.
[0026] In some instances, the upright surface cleaning device 400 may include a controller 416 and one or more sensors 418 communicatively coupled to the controller 416. The controller 416 may be configured to control operation of one or more of the suction motor 410 and / or the agitator 412 based, at least in part, on an output generated by the one or more sensors 418. In some instances, the one or more sensors 418 include a surface type sensor 420. In some instances, the surface type sensor 420 may be an acoustic sensor configured to detect sounds within an environment (e.g., sounds within the human audible spectrum). In these instances, sounds generated as a result of operation of the upright surface cleaning device 400 (e.g., from the suction motor 410 and / or the agitator 412) may be reflected from the surface to be cleaned 414 and detected by the surface type sensor 420. Based, at least in part, on changes in output from the surface type sensor 420 resulting from the detected sounds, the controller 416 may be configured to determine a surface type. In this way, the surface type sensor 420 may be able to omit and acoustic emitter and only have an acoustic detector (e.g., a microphone).
[0027] The surface type sensor 420 may be coupled to the surface cleaning head 402 such that the surface type sensor 420 is able to detect sounds generated by operation of the upright surface cleaning device 400 within an environment. For example, the surface type sensor 420 may be coupled to the surface cleaning head 402 proximate the agitator 412. By way of further example, the surface type sensor 420 may be coupled to the surface cleaning head 402 such that the surface type sensor 420 faces the surface to be cleaned 414. By way of still further example, the surface type sensor 420 may be coupled to a room facing surface of the surface cleaning head 402.
[0028] FIG. 5 shows a schematic example of a surface type detection system 500. The surface type detection system 500 may be used in conjunction with a surface cleaning device such as the surface cleaning device 100 of FIG. 1 (e.g., the upright or robotic surface cleaning device 200 or 400). As shown, the surface type detection system 500 includes a microphone 502 configured to detect sounds within a surrounding environment (e.g., sounds generated by operation of a suction motor, such as suction motor 108, 208, or 410, and / or by operation of an agitator, such as agitator 110, 218, or 412) and a controller 504 (e.g., the controller 116, 216, or 416 of FIGS. 1, 2 and 4, respectively), the controller 504 being communicatively coupled to the microphone 502. The microphone 502 is configured to generate a microphone output that corresponds to the detected sounds, the microphone output being received by the controller 504. The microphone output changes based on a surface type being traversed (e.g., as a result of a change in load experienced by the suction motor, the agitator motor, the side brush motor, and / or other component of the surface cleaning device, as a result of contact between the agitator, the side brush, and / or any other component of the surface cleaning device with the surface to be cleaned, as a result of acoustic absorption properties of the surface type, and / or as a result of changes in airflow entering the surface cleaning device). As such, the microphone output can be used to determine a surface type. The controller 504 may be configured to control the behaviors of one or more components of a surface cleaning device (e.g., the surface cleaning devices 100, 200, and / or 400). For example, the controller 504 may adjust one or more behaviors, based at least in part, on the microphone output. In this example, the controller 504 may include a machine learning model configured to determine a surface type based, at least in part, on the microphone output and the controller 504 may adjust one or more behaviors based, at least in part, on the determined surface type.
[0029] The microphone 502 may be configured to detect sounds at least within the human audible range (e.g., 20 hertz to 20 kilo-hertz). For example, a frequency range of interest may extend from about 800 hertz (Hz) to about 10,000 Hz. In this example, the sample rate may be about 20,000 Hz. By way of further example, a frequency range of interest may extend, for example, from about 20 Hz to about 20,000 Hz. In this example, the sample rate may be about 44,000 Hz. Higher frequency sounds may experience greater dampening on soft materials than hard materials.
[0030] The detected sounds may include sounds generated as a result of operation of the surface cleaning device. For example, an operation of a suction motor generates sound that may be detected by the microphone 502. Prior to being detected by the microphone 502, the sound generated by the suction motor may be reflected from one or more surfaces (e.g., a floor) within the surrounding environment. Reflection of the sound from the one or more surfaces alters the sound, relative to a baseline sound generated at the suction motor. The controller 504 may be configured to analyze the output generated by the microphone 502 resulting from the reflected sound and determine a surface type based, at least in part, on the alterations to the sound caused by the reflections within the surrounding environment. In another example, rotation of an agitator (e.g., brush roll) generates a sound that may be detected by the microphone 502 (e.g., as a result of the agitator contacting a surface to be cleaned). Sounds resulting from an impact of the agitator against the surface to be cleaned may change based on a surface type of the surface to be cleaned. The controller 504 may be configured to analyze the output of the microphone generated by the sound resulting from the agitator impacting the surface to be cleaned and determine a surface type based on the output. In other words, the microphone 502 can be configured to detect sounds generated by one or more components of the surface cleaning device (in addition to any ambient sounds) and the controller 504 can be configured to analyze the output of the microphone 502 corresponding to the detected sounds (both ambient and generated) to detect patterns in order to determine a surface type, wherein the detected sound changes as a result of changes in surface type being cleaned. The microphone 502 may be a Micro-Electro-Mechanical System (MEMS) microphone, a piezo microphone, or an electret microphone.
[0031] The controller 504 may operate according to a machine learning model. The machine learning model may be generated through a training method in which a machine learning algorithm is used to recognize patterns within output from the microphone 502. For example, when the surface cleaning device is operated on a carpeted surface, the sounds generated as a result of the operation may result in an output from the microphone 502 corresponding to a carpet pattern. By way of further example, when the surface cleaning device is operated on a hard surface, the sounds generated as a result of the operation may result in an output from the microphone 502 corresponding to a hard surface pattern. In some instances, the machine learning algorithm may be used to recognize patterns within an output of the microphone 502 corresponding to additional surface types. For example, the machine learning algorithm may be used to recognize patterns corresponding to one or more of shag carpet, high-pile carpet, low-pile carpet, level loop carpet, plush carpet, mats (e.g., rubber bath mats, fatigue mats, welcome mats, and / or the like), hardwood, tile, vinyl, concrete (bare or painted), and / or any other surface type.
[0032] While the microphone 502 is generally described as being configured to detect sounds generated by the surface cleaning device, other configurations are possible. For example, the microphone 502 may be able to receive voice commands from a user and the controller 504 may be configured to cause changes in the behavior of the surface cleaning device based, at least in part, on the voice commands.
[0033] FIG. 6 is a flow chart corresponding to a training method 600 for using a machine learning algorithm to generate a machine learning model for use with the controller 116, 216, 416, and / or 504 of FIGS. 1, 2, 4, and 5 to determine a surface type. The machine learning model may then be used during cleaning operations to determine a surface type. In some instances, the machine learning model may be used with multiple surface cleaning devices (e.g., having a configuration substantially the same as the surface cleaning machine used to generate the machine learning model). One or more portions of the training method 600 may be embodied as one or more instructions stored in one or more memories, wherein the one or more instructions are configured to be executed by one or more processors. Additionally, or alternatively, one or more portions of the training method 600 may be carried out in any combination of software, firmware, or circuitry (e.g., an application-specific integrated circuit).
[0034] The training method 600 may include a step 602. The step 602 includes operating a surface cleaning device (e.g., the surface cleaning device 100, 200, and / or 400) while on a surface having a known surface type (e.g., a carpeted surface type, a hard surface type, etc.). The surface cleaning device may be operated on a given surface type for a test period. A length of the test period may be in a range of, for example, 10 seconds to 90 seconds. By way of further example, a length of the test period may be in a range of 20 second to 70 seconds. By way of still further example, a length of the test period may be about (e.g., within 10% of, 5%, of, 4% of, 3% of, 2% of, or 1% of) 50 seconds.
[0035] During the step 602, a surface type detection system (e.g., the surface type detection system 500 of FIG. 5) coupled to the surface cleaning device is configured to detect sound generated by operation of the surface cleaning device. The surface type detection system includes a microphone (e.g., the microphone 502) connected to a controller (e.g., the controller 504). While the surface cleaning device is being operated for the test period, the microphone is configured to generate an output that is recorded by the controller. The generated output corresponds to the sound generated by the surface cleaning device during a test period. The recorded output may be generally referred to as a sound sample.
[0036] After recording a first sound sample, additional sound samples can be recorded, wherein each sound sample is recorded during a new test period. For example, between 10 and 150 sound samples may be recorded. By way of further example, between 30 and 100 sound samples may be recorded. By way of still further example, about 50 sound samples may be recorded.
[0037] The step 602 may be repeated for multiple surface types to generate multiple surface type samples, each surface type sample being associated with corresponding sound samples. For example, the step 602 may be repeated to generate at least three surface type samples (e.g., a hard surface type sample, a shag carpet surface type sample, and a low pile carpet surface type sample), each of the at least three surface type samples having one or more corresponding sound samples (e.g., each surface type sample may be associated with about 50 sound samples for a given surface type).
[0038] The training method 600 may include a step 604. The step 604 includes dividing each sound sample into at least two sound segments. Each sound segment may correspond to a predetermined segment length, the segment length being a period of time that is less than the test period. For example, each segment length may be in a range of 0.25 second to 20 seconds. By way of further example, each segment length may be in a range of 0.5 second to 5 seconds. By way of still further example, each segment length may be about 1 second. By way of still further example, when the test period is 50 seconds, each sound sample may be divided into 50 sound segments having a 1 second segment length.
[0039] The training method 600 may include a step 606. The step 606 includes determining octave bands for each sound segment to generate sound features. Computing the octave bands may include converting the output from the microphone into frequency data (e.g., using a Fourier transform such as a discrete or fast Fourier transform). The octave bands may be one-third (⅓) octave bands. The octave bands may be A-weighted. Using an A-weighting deemphasizes lower frequency noise. In a residential setting, a significant portion of background noise may be lower frequencies and use of an A-weighting may improve a signal-to-noise ratio by deemphasizing the lower frequencies. For each octave band, a frequency content of the octave band may be averaged (or, in some instances, summed) together to obtain a single number (e.g., a sound feature) corresponding to that octave band. In one example, for each sound segment (e.g., having a 1 second segment length) 39 A-weighted one-third octave bands may be determined to generate 39 sound features.
[0040] The training method 600 may include a step 608. The step 608 includes providing each sound feature to a machine learning algorithm to generate a machine learning model. For example, each sound feature for each sound sample may be organized into a matrix capable of being provided to the machine learning algorithm. The machine learning algorithm analyzes the matrix to determine surface type patterns corresponding to a specific surface type. As such, in some instances, the machine learning algorithm may be generally referred to as a supervised machine learning algorithm. The machine learning algorithm may be a Support Vector Machine (SVM) algorithm, a k-Nearest Neighbor (k-NN) algorithm, a decision tree regressor algorithm, binary classifier, and / or any other machine learning algorithm.
[0041] FIG. 7 shows an example operation method 700 of a surface cleaning device that operates according to a machine learning model configured to determine a surface type (e.g., as generated during the training method of FIG. 6). One or more portions of the operation method 700 may be embodied as one or more instructions stored in one or more memories, wherein the one or more instructions are configured to be executed by one or more processors (e.g., one or memories and / or one or more processors of the controller 116, 216, 416, or 504). Additionally, or alternatively, one or more portions of the operation method 700 may be carried out in any combination of software, firmware, or circuitry (e.g., an application-specific integrated circuit).
[0042] As shown, the operation method 700 may include a step 702. The step 702 includes monitoring an output of a microphone (e.g., the microphone 502 of FIG. 5) of a surface cleaning device as the surface cleaning device moves across a surface to be cleaned. The step 702 may further include monitoring an output of one or more additional sensors.
[0043] The output of one or more additional sensors may be used to improve a confidence in a surface type determined pursuant to the machine learning model. For example, the one or more sensors may include a motor current sensor configured to detect a current draw as a result of operation of a suction motor and / or an agitator (e.g., a brush roll). The current draw resulting from operation of the suction motor and / or rotation of the agitator may change based on surface type. If the surface type corresponding to the current draw is consistent with the surface type determined using the machine learning model, a confidence in the determined surface type may be improved. By way of further example, the one or more additional sensors may include an inertial measurement unit (IMU). In this example, when there is a change in surface type, an output of the IMU may change (e.g., as a result of crossing over a threshold between surface types). When the IMU output indicates a change in surface type occurred, a confidence in the surface type determined by the machine learning model may be improved if the machine learning model also detects a change in surface type.
[0044] The output of one or more additional sensors may be configured to detect a status of one or more components of the surface cleaning device that may have a detrimental impact on the accuracy of the machine learning model. For example, one or more sensors may be provided to detect and / or estimate an amount of wear of an agitator (e.g., a brush roll). In this example, as the agitator wears, the sound generated as a result of the agitator contacting a surface to be cleaned may change. As such, in some instances, the machine learning model may be adjusted to account for the wear and / or a notification may be provided to the user that the agitator should be changed. In these instances, a worn-agitator machine learning model may be used instead of the initial machine learning model until the worn-agitator is replaced. By way of further example, one or more dust cup sensors may be provided to detect a quantity of debris within the dust cup. In this example, as debris accumulates in the dust cup, a flow restriction may be created, causing a sound of the suction motor to change. As such, in some instances, the machine learning model may be adjusted to account acoustic changes resulting from debris accumulation. By way of still further example, one or more dust cup sensors may be provided to detect a debris type (e.g., hard or soft) within the dust cup. In this example, when the dust cup primarily includes hard debris, noise may be generated as result of the hard debris moving within and impacting the dust cup. As such, in some instances, the machine learning model may be adjusted to account for acoustic changes resulting from hard debris moving within and impacting the dust cup.
[0045] In some instances, the controller may be configured to perform a self-diagnosis of the microphone (e.g., if a determined confidence level in a determined surface type is below a threshold). For example, when the microphone becomes obstructed (e.g., by debris), the microphone may not reliably detect generated sound. In this example, an obstruction alert may be generated and transmitted to a user.
[0046] As shown, the operation method 700 may include a step 704. The step 704 includes analyzing the output of the microphone and determining a surface type using a machine learning model (e.g., as generated during the training method 600 of FIG. 6). The machine learning model may be configured to identify at least 2, at least 3, at least 4, at least 5, at least 6, and / or any other number of surface types. For example, the machine learning model may be configured to identify at least one hard surface type and at least two soft surface types (e.g., low pile carpet and shag carpet). By way of further example, the machine learning model may be configured to identify at least two hard surface types (e.g., tile, hardwood, vinyl, and / or the like) and at least two soft surface types (e.g., low pile carpet, shag carpet, and / or the like).
[0047] In one example, analyzing the output of the microphone and determining a surface type with the machine learning model may include determining octave bands for one or more sound segments to generate sound features. The octave bands may be generated by converting the output from the microphone into frequency data (e.g., using a Fourier transform such as a discrete or fast Fourier transform). The sound segments may correspond to predetermined time intervals of operation (e.g., about 1 second, about 2 seconds, about 3 seconds, about 10 seconds, and / or any other predetermined time interval). The octave bands may be A-weighted and / or one-third octave bands. The frequency content of each of the octave bands may be averaged (or, in some instances, summed) together to obtain a sound feature corresponding to that octave band. The sound features may then be provided to the machine learning model to determine a surface type. In some instances, multiple sound segments may be analyzed before making a determination of surface type (e.g., the determination of surface type may be made after a number or ratio of analyzed sound segments that indicate a common surface type exceed a threshold). When the machine learning model is based on a decision tree regressor, each sound feature may be associated with a tree, the value of each sound feature may propagate down the tree until reaching a final leaf node, and a final values associated with each tree may then be added together and divided by the number of trees to get an output. The output of decision tree regressor may be between 0 and 1. As should be appreciated, the process of generating the sound features for input to the machine learning model is similar to the process described in the training method 600 of FIG. 6 for generating the machine learning model (e.g., as discussed in step 606). In other words, the inputs provided to the machine learning model are generated in the same manner as the inputs provided to the machine learning algorithm for generation of the machine learning model.
[0048] As shown, the operation method may include a step 706. The step 706 includes adjusting a behavior of the surface cleaning device (e.g., surface cleaning device 100, 200, or 400) based, at least in part, on the determined surface type. For example, in the case of the robotic surface cleaning device 200 of FIG. 2, the robotic surface cleaning device 200 may be configured to adjust its ride height and / or a type of cleaning (e.g., wet cleaning or dry cleaning) based, at least in part, on the surface type. By way of further example, in the case of the robotic surface cleaning device 200, when the machine learning model is configured to differentiate between at least two hard surface types (e.g., hardwood and tile), the robotic surface cleaning device 200 may carry out a specific hard surface behavior (e.g., wet or dry mopping) assigned to the detected hard surface type. By way of still further example, in the case of the upright surface cleaning device 400, the upright surface cleaning device 400 may be configured to adjust a suction force and / or a rotational speed of the agitator 412 based, at least in part, on the determined surface type. After adjusting the behavior of the surface cleaning device, the method may return to the step 702 to monitor for a change in surface type. While the examples are described in relation to the robotic surface cleaning device 200 or the upright surface cleaning device 400, such description is made for the sake of clarity. As such, behaviors described in relation to the robotic surface cleaning device 200 may be implemented in the upright surface cleaning device 400 and vice-versa.
[0049] FIG. 8 shows a perspective view of a surface cleaning head 800, which is an example of the surface cleaning head 402 of FIG. 4. As shown, the surface cleaning head 800 includes a head body 802, a first and a second brush roll 804 and 806 rotatably coupled to the head body 802, and a microphone 808 coupled to the head body 802. As shown, the head body 802 includes at least one agitator chamber 810 configured to receive at least a portion of one or more the first and / or second brush rolls 804 and / or 806. As shown, the microphone 808 is at least partially received within the agitator chamber 810. Such a configuration may provide isolation of the microphone 808 from at least a portion of any ambient noise.
[0050] FIG. 9 shows a perspective view of a surface cleaning head 900, which is an example of the surface cleaning head 402 of FIG. 4. As shown, the surface cleaning head 900 includes a head body 902, a first and a second brush roll 904 and 906 rotatably coupled to the head body 902, and a microphone 908 coupled to the head body 902. As shown, the head body 902 includes at least one agitator chamber 910 configured to receive at least a portion of one or more the first and / or second brush rolls 904 and / or 906. As shown, the microphone 908 is external to the agitator chamber 910. Such a configuration may, while potentially exposing the microphone 908 to ambient noise, increase an ability of the microphone 908 to detect the sounds generated during operation, relative to the placement of shown in FIG. 8. In some instances, a surface cleaning head may include two microphones, wherein a first is positioned as shown in FIG. 8 and a second is positioned as shown in FIG. 9. In these instances, the outputs of both microphones may be analyzed (e.g., to reduce the impact of ambient noise).
[0051] FIG. 10 shows a bottom view of a robotic surface cleaning device 1000, which is an example of the robotic surface cleaning device 200 of FIG. 2. As shown, the robotic surface cleaning device 1000 includes a main body 1002, a side brush 1004 rotatable relative to the main body 1002, an agitator 1006 rotatable relative to the main body 1002, a plurality of driven wheels 1008, and a microphone 1010. As shown, the microphone 1010 is downward facing (i.e., towards a surface to be cleaned) and positioned forward of the plurality of driven wheels 1008 (relative to a forward direction of movement). In some instances, a plurality of microphones 1010 may be provided, wherein the microphones 1010 are on opposing sides of a forward-backward axis 1012. The forward-backward axis 1012 extends perpendicular to a rotation axis 1014 of the plurality of driven wheels 1008.
[0052] As also shown, the robotic surface cleaning device 1000 may include a mopping assembly 1016 removably coupled to the main body 1002. The mopping assembly 1016 may be configured for dry mopping (e.g., dusting) and / or wet mopping. The mopping assembly 1016 may include a mop pad 1018 coupled to a mop tray 1020.
[0053] An example of a surface cleaning device, consistent with the present disclosure, may include a suction motor, an agitator, a microphone configured to detect sound generated by the suction motor and the agitator and to generate a microphone output corresponding to the detected sound, and a controller configured to control rotation of the suction motor and rotation of the agitator and communicatively coupled to the microphone, the controller including a machine learning model configured to determine a surface type based, at least in part, on the microphone output.
[0054] In some instances, the machine learning model may be configured to determine at least three surface types. In some instances, the machine learning model may be configured to determine at least two surface types. In some instances, the machine learning model may be generated using a Support Vector Machine (SVM) machine learning algorithm. In some instances, the machine learning model may be generated using a k-Nearest Neighbor (k-NN) machine learning algorithm. In some instances, the machine learning model may be generated using a decision tree regressor machine learning algorithm. In some instances, the machine learning model may be generated using a supervised machine learning algorithm. In some instances, the controller may be configured to adjust a rotational speed of the agitator based, at least in part, on the determined surface type. In some instances, the controller may be configured to adjust a rotational speed of the suction motor based, at least in part, on the determined surface type. In some instances, the controller may be configured to cause a ride height of the surface cleaning device to be adjusted based, at least in part, on the determined surface type. In some instances, the machine learning model may be configured to determine at least two soft surface types and at least one hard surface type.
[0055] An example of a method for generating a machine learning model configured to determine a surface type, consistent with the present disclosure, may include operating a surface cleaning device on a surface having a known surface type for a test period, during operation of the surface cleaning device during the test period, detecting, using a microphone, sound generated due to the operation of the surface cleaning device, generating, using an output of the microphone, a sound sample, dividing the sound sample into at least two sound segments, each sound segment having a segment length that is less than the test period, determining octave bands for each sound segment to generate sound features, and providing the sound features to a machine learning algorithm to generate the machine learning model.
[0056] In some instances, the octave bands may be one-third octave bands. In some instances, the octave bands may be A-weighted. In some instances, test period may be about 50 seconds. In some instances, the segment length may be about 1 second. In some instances, the machine learning algorithm may be a Support Vector Machine (SVM) machine learning algorithm. In some instances, the machine learning algorithm may be a k-Nearest Neighbor (k-NN) machine learning algorithm. In some instances, the machine learning algorithm may be a decision tree regressor machine learning algorithm. In some instances, a frequency content for each octave band may be averaged to generate a corresponding sound feature. In some instances, for each sound segment, 39 A-weighted one-third octave bands may be determined to generate 39 sound features.
[0057] Another example of a surface cleaning device, consistent with the present disclosure, may include a suction motor, an agitator, a microphone configured to detect sound generated by the suction motor and the agitator and to generate a microphone output corresponding to the detected sound, and a controller configured to control rotation of the suction motor and rotation of the agitator and communicatively coupled to the microphone, the controller including a machine learning model configured to determine a surface type based, at least in part, on the microphone output. The controller may be further configured to generate a sound sample from the output of the microphone, divide the sound sample into at least two sound segments, determine octave bands for each sound segment to generate sound features, and provide the sound features to the machine learning model to determine the surface type.
[0058] While the principles of the invention have been described herein, it is to be understood by those skilled in the art that this description is made only by way of example and not as a limitation as to the scope of the invention. Other embodiments are contemplated within the scope of the present invention in addition to the exemplary embodiments shown and described herein. Modifications and substitutions by one of ordinary skill in the art are considered to be within the scope of the present invention, which is not to be limited except by the following claims.
Claims
1. A surface cleaning device comprising:a suction motor;an agitator;a microphone configured to detect sound generated by the suction motor and the agitator and to generate a microphone output corresponding to the detected sound; anda controller configured to control rotation of the suction motor and rotation of the agitator and communicatively coupled to the microphone, the controller including a machine learning model configured to determine a surface type based, at least in part, on the microphone output.
2. The surface cleaning device of claim 1, wherein the machine learning model is configured to determine at least two surface types.
3. The surface cleaning device of claim 1, wherein the machine learning model is generated using a Support Vector Machine (SVM) machine learning algorithm.
4. The surface cleaning device of claim 1, wherein the machine learning model is generated using a k-Nearest Neighbor (k-NN) machine learning algorithm.
5. The surface cleaning device of claim 1, wherein the machine learning model is generated using a decision tree regressor machine learning algorithm.
6. The surface cleaning device of claim 1, wherein the machine learning model is generated using a supervised machine learning algorithm.
7. The surface cleaning device of claim 1, wherein the controller is configured to adjust a rotational speed of the agitator based, at least in part, on the determined surface type.
8. The surface cleaning device of claim 1, wherein the controller is configured to adjust a rotational speed of the suction motor based, at least in part, on the determined surface type.
9. The surface cleaning device of claim 1, wherein the controller is configured to cause a ride height of the surface cleaning device to be adjusted based, at least in part, on the determined surface type.
10. The surface cleaning device of claim 1, wherein the machine learning model is configured to determine at least two soft surface types and at least one hard surface type.
11. A method for generating a machine learning model configured to determine a surface type comprising:operating a surface cleaning device on a surface having a known surface type for a test period;during operation of the surface cleaning device during the test period, detecting, using a microphone, sound generated due to the operation of the surface cleaning device;generating, using an output of the microphone, a sound sample;dividing the sound sample into at least two sound segments, each sound segment having a segment length that is less than the test period;determining octave bands for each sound segment to generate sound features; andproviding the sound features to a machine learning algorithm to generate the machine learning model.
12. The method of claim 11, wherein the octave bands are one-third octave bands.
13. The method of claim 11, wherein the octave bands are A-weighted.
14. The method of claim 11, wherein the test period is about 50 seconds.
15. The method of claim 14, wherein the segment length is about 1 second.
16. The method of claim 11, wherein the machine learning algorithm is a Support Vector Machine (SVM) machine learning algorithm.
17. The method of claim 11, wherein the machine learning algorithm is a k-Nearest Neighbor (k-NN) machine learning algorithm.
18. The method of claim 11, wherein the machine learning algorithm is a decision tree regressor machine learning algorithm.
19. The method of claim 11, wherein a frequency content for each octave band is averaged to generate a corresponding sound feature.
20. A surface cleaning device comprising:a suction motor;an agitator;a microphone configured to detect sound generated by the suction motor and the agitator and to generate a microphone output corresponding to the detected sound; anda controller configured to control rotation of the suction motor and rotation of the agitator and communicatively coupled to the microphone, the controller including a machine learning model configured to determine a surface type based, at least in part, on the microphone output, wherein the controller is further configured to:generate a sound sample from the output of the microphone;divide the sound sample into at least two sound segments;determine octave bands for each sound segment to generate sound features; andprovide the sound features to the machine learning model to determine the surface type.
