US-20260124086-A1 - BED HAVING PRESENCE DETECTING FEATURE

US20260124086A1US 20260124086 A1US20260124086 A1US 20260124086A1US-20260124086-A1

Abstract

A first controller is in data communication with a first pressure sensor and is configured to receive first pressure readings, and transmit the first pressure readings. The system further includes a second controller in data communication with the second pressure sensor, the controller configured to receive one or more presence classifiers. The second controller is further configured to run the received presence classifiers on second pressure readings in order to collect one or more presence votes from the running presence classifiers. The second controller is further configured to determine, from the one or more presence votes, a presence state of a user on the second bed. The second controller is further configured to responsive to the determined presence state; operate the bed system according to the determined presence state.

Inventors

Omid Sayadi
Ramazan Demirli
Shruthi Balasubramanian
Farzad Siyahjani

Assignees

SLEEP NUMBER CORPORATION

Dates

Publication Date: 20260507
Application Date: 20251231

Claims (20)

1 - 20 . (canceled)
21 . A system comprising: a controller configured to be in data communication with one or more sensors of a bed system, the controller configured to: retrieve one or more presence classifiers, wherein the one or more presence classifiers were trained, using machine learning techniques and based on training data, to generate a respective presence vote when run by the controller on sensor readings; receive sensor readings from the one or more sensors; run the one or more presence classifiers on the received sensor readings in order to collect one or more presence votes from the running presence classifiers; determine, from the one or more presence votes, a presence state of a user on the bed system; and operate the bed system according to the determined presence state.
22 . The system of claim 21 , wherein the training data comprises sensor readings collected at one or more sensors of a second bed system.
23 . The system of claim 21 , wherein the system further comprises a pad configured to be used with the bed system.
24 . The system of claim 23 , wherein the pad is configured to be placed on top of a mattress of the bed system.
25 . The system of claim 23 , wherein the pad includes an activatable heating element.
26 . The system of claim 23 , wherein the pad includes an activatable temperature modulation element.
27 . The system of claim 26 , wherein the activatable temperature modulation element is configured to i) actively heat and ii) actively cool.
28 . The system of claim 21 , wherein the one or more presence classifiers were trained at a remote server configured to receive the training data over a data network.
29 . The system of claim 21 , wherein the one or more presence classifiers were trained, using the machine learning techniques and based on the training data, by: generating feature sets from the training data; mapping the training data to a kernel space; and training the one or more presence classifiers with the feature sets so that, based on the training data in the kernel space, the one or more presence classifiers are able to classify unseen data.
30 . The system of claim 21 , wherein the received sensor readings include pressure readings.
31 . A bed system comprising: a mattress; one or more sensors; and a controller configured to be in data communication with the one or more sensors, the controller being configured to: retrieve one or more presence classifiers, wherein the one or more presence classifiers were trained, using machine learning techniques and based on training data, to generate a respective presence vote when run by the controller on sensor readings; receive sensor readings from the one or more sensors; run the one or more presence classifiers on the received sensor readings in order to collect one or more presence votes from the running presence classifiers; determine, from the one or more presence votes, a presence state of a user on the bed system; and operate the bed system according to the determined presence state.
32 . A system comprising: a pad configured to be placed on top of a mattress of a bed system; one or more sensors; and a controller configured to be in data communication with the one or more sensors, the controller being configured to: retrieve one or more presence classifiers, wherein the one or more presence classifiers were trained, using machine learning techniques and based on training data, to generate a respective presence vote when run by the controller on sensor readings; receive sensor readings from the one or more sensors; run the one or more presence classifiers on the received sensor readings in order to collect one or more presence votes from the running presence classifiers; determine, from the one or more presence votes, a presence state of a user on the bed system; and operate the bed system according to the determined presence state.
33 . A bed system for presence mitigation, the bed system comprising: a mattress; one or more sensors configured to generate one or more data streams indicative of a physiological state of a specific user on the mattress; at least one physical adjustment mechanism configured to change a physical configuration of the bed system; and a controller including one or more processors and memory, the controller in data communication with the one or more sensors and the at least one physical adjustment mechanism, the controller configured to: execute a trained presence determination model, wherein the trained presence determination model is a deep neural network that was trained using historical sensor data collected from a population of users other than the specific user; process, using the trained presence determination model, the data streams of the specific user to determine a presence state of the specific user; determine whether the presence state indicates that the specific user is present on the mattress; and in response to determining the presence state indicates that the specific user is present on the mattress, cause the at least one physical adjustment mechanism to change the physical configuration of the bed system.
34 . The bed system of claim 33 , wherein the at least one physical adjustment mechanism is an articulation mechanism, and wherein changing the physical configuration of the bed system comprises adjusting an incline position of a head portion of the mattress.
35 . The bed system of claim 33 , wherein determining the presence state is further based on a confidence score generated by the trained presence determination model, and wherein the controller causes the at least one physical adjustment mechanism only when the confidence score exceeds a predetermined threshold.
36 . A bed system for presence mitigation, the bed system comprising: a mattress; one or more sensors configured to generate one or more data streams indicative of a physiological state of a user on the mattress; at least one physical adjustment mechanism; and a controller in data communication with the one or more sensors and the at least one physical adjustment mechanism, the controller configured to: process the one or more data streams using a plurality of machine learning models to generate a plurality of independent presence votes, wherein each machine learning model in the plurality of machine learning models is configured to generate a respective independent presence vote; aggregate, using a vote-counting scheme, the independent presence votes from the plurality of machine learning models to determine a final presence state; determine whether the final presence state indicates that the user is present on the mattress; and in response to determining that the final presence state indicates that the user is present on the mattress, cause the at least one physical adjustment mechanism to change a physical configuration of the bed system.
37 . The bed system of claim 36 , wherein the vote-counting scheme comprises for each independent presence vote of the independent presence votes from each machine learning model of the machine learning models: applying a distinct weight to the independent presence vote from the machine learning model based on a historical accuracy of the machine learning model.
38 . A bed system for reliable presence mitigation, the bed system comprising: a mattress; one or more sensors configured to generate one or more data streams from sensing a user on the mattress; at least one physical adjustment mechanism; and a controller in data communication with the one or more sensors and the at least one physical adjustment mechanism, the controller configured to: execute a machine learning model to process the one or more data streams; generate, over a series of successive time windows, a corresponding series of individual confidence scores, each individual confidence score indicating a likelihood of being present on the mattress within its respective time window; aggregate the corresponding series of individual confidence scores from the successive time windows into an aggregated confidence score; determine whether the aggregated confidence score is greater than a predetermined reliability threshold; and in response to determining that the aggregated confidence score is greater than the predetermined reliability threshold, cause the at least one physical adjustment mechanism to change a physical configuration of the bed system.
39 . A remote server system for deploying presence detection models, the remote server system comprising: a network interface configured to communicate with a fleet of bed systems over a network; one or more processors; and a non-transitory computer-readable medium storing instructions that, when executed by the one or more processors, cause the remote server system to: receive, via the network interface from the fleet of bed systems, historical sensor data indicative of physiological states of a plurality of users; train a deep neural network model using the received historical sensor data from the plurality of users to generate a trained presence detection model capable of determining a presence state from real-time sensor data; and transmit, via the network interface, the trained presence detection model to at least one bed system in the fleet, thereby enabling the at least one bed system to run the trained presence detection model on a controller of the bed system.

Description

CROSS-REFERENCE TO RELATED APPLICATIONS This application is a continuation of U.S. application Ser. No. 16/233,230, filed Dec. 27, 2018, which claims priority to U.S. Application Ser. No. 62/611,050, filed on Dec. 28, 2017. The disclosure of the prior application is considered part of the disclosure of this application, and is incorporated in its entirety into this application. BACKGROUND In general, a bed is a piece of furniture used as a location to sleep or relax. Many modern beds include a soft mattress on a bed frame. The mattress may include springs, foam material, and/or an air chamber to support the weight of one or more occupants. SUMMARY In one aspect, a bed system includes a first bed that includes a first mattress. The system further includes a first pressure sensor in communication with the first mattress to sense pressure applied to the first mattress. The system further includes a first controller in data communication with the first pressure sensor, the first controller configured to receive, from the first pressure sensor, first pressure readings indicative of the sensed pressure of the inflatable chamber. The first controller is configured to transmit the first pressure readings to a remote server such that the remote server is able to generate one or more presence classifiers using training data created with the first pressure readings such that, when the one or more presence classifiers are run by a controller on incoming pressure readings, the one or more presence classifiers provides a presence vote. The system further includes a second bed that includes a second mattress. The system further includes a second pressure sensor in communication with the second mattress to sense pressure applied to the second mattress. The system further includes a second controller in data communication with the second pressure sensor, the controller configured to receive the one or more presence classifiers. The second controller is further configured to run the received presence classifiers on second pressure readings in order to collect one or more presence votes from the running presence classifiers. The second controller is further configured to determine, from the one or more presence votes, a presence state of a user on the second bed. The second controller is further configured to responsive to the determined presence state, operate the bed system according to the determined presence state. Other systems, devices, methods, and computer-readable media may be used. Implementations can include any, all, or none of the following features. Operating the bed system according to the determined presence state includes one of the list consisting of turning on a light, turning off a light, turning on a warming feature, changing firmness of the mattress, and articulating a foundation of the bed system. The bed system including the remote server. The remote server is physically remote from the first controller and the second controller; and wherein the remote server is in data communication with the first controller and the second controller. The remote server is configured to generate, from the training data, the one or more presence classifiers; and send, to the second controller, the one or more presence classifiers. Generating, from the training data, the one or more presence classifiers includes generating a feature set from the training data; mapping the training data to a kernel space; and training a classifier with the feature set so that, based on the training data in kernel space, the classifier is able to classify unseen data. Training a classifier includes unsupervised training. The unsupervised training includes at least one of the group including k-means clustering, mixture modeling, hierarchical clustering, self-organizing mapping, and hidden Markov modelling. Training a classifier includes supervised training. The supervised training includes providing the remote server with a set of annotations for the training data. The annotations for the training data are provided by a human. The annotations for the training data are provided programmatically. Generating the one or more presence classifiers includes training a deep learning model on the training data; Training the deep learning model on the training data includes generating an initial neural network configured to receive pressure data and generate presence votes. The presence vote includes a presence classification and a confidence value. Generating the one or more presence classifiers includes determining a loss value for the initial neural network; and iteratively refining, beginning with the initial neural network, to a final neural network having a lower loss value than the initial neural network. The iterative refining is performed with a gradient descent process until a lower loss value cannot be found with the gradient descent process. A particular presence classifier is used for multiple users in multiple beds. The presence classif