System and method for improving sleep quality

Abstract

A method of controlling a sleep enhancement device or system (11) for improving sleep quality of a sleeper (12), the method comprising the steps of: estimating a sleep stage in which the sleeper is based on information from at least one or more sensors, said sleep stage being selected from sleep stage 3 (N3) and sleep stage 4 (REM) or another sleep stage; and when the sleeper is estimated to be in a sleep stage 3 (N3) and / or when the sleeper is estimated to be in a sleep stage 4 (REM), performing one of the following steps: activating a sleep enhancement device or system to begin providing an input acting on the sleeper; stopping the sleep enhancement device or system to stop providing the input acting on the sleeper; a sleep enhancement device or system is adjusted to change the level of input acting on the sleeper.

Description

Technical Field

[0001] The present invention generally relates to a method of controlling a sleep enhancement device or system for improving the sleep quality of a sleeper, a system for implementing the method, and a method of generating control data for controlling a sleep enhancement device or system to improve sleep quality through a machine learning model. Background Art

[0002] Sleep is vital for everyone, but it is not always easy to get. A person may have trouble falling asleep, staying asleep, waking up, returning to sleep after waking up, waking up and still feeling tired, or any combination of these.

[0003] For example, sleep disruption may occur occasionally due to illness or snoring, but sleep disruption due to a sleep disorder can be extremely debilitating and may affect every aspect of a person's life. There are more than 80 different known sleep disorders.

[0004] Some people may not have sleep problems per se, but may have busy personal and / or work lives that place a lot of demands on their time. Most people need 7-8 hours of sleep per day but may often fail to achieve this, leaving them with a sleep debt that can accumulate and become debilitating in some way.

[0005] Somnology is the scientific study of sleep and can be used to help people who are experiencing disrupted sleep, particularly in relation to sleep disorders. Somnology divides sleep into four distinct stages / periods, referred to as stage 1 (N1), stage 2 (N2), stage 3 (N3), and stage 4 (REM), which are discussed further below.

[0006] During sleep, the sleep stages from stage 1 (N1) to sleep stage 4 (REM) are progressive, which forms a sleep cycle, after which the cycle returns from sleep stage 2 to 4. A complete sleep cycle for an adult is 90 to 110 minutes. The first or initial sleep cycle consists of a relatively short REM (rapid eye movement) sleep and a relatively long deep sleep. During the night, the time of REM sleep increases, while the time of deep sleep decreases. A person who sleeps throughout the night usually goes through four to six sleep cycles, which means they go through different sleep stages many times.

[0007] Considering the sleep stages in more detail, the first sleep stage is sleep stage 1, which is considered "light" sleep where the body is not yet fully relaxed. Sleep stage 1 (N1) can last about 5 to 10 minutes and represents the beginning of the transition from wakefulness to sleep.

[0008] The next sleep stage 2 (N2) can last from 10 to 25 minutes in the first sleep cycle (i.e. the first time you enter sleep stage 2 after falling asleep). Each sleep stage 2 gradually gets longer as a person sleeps. In general, a person usually spends about 50% of their total sleep time in sleep stage 2. In sleep stage 2, eye movement stops and brain waves slow down.

[0009] Sleep stage 3 (N3) is deep sleep, accounting for about 20% of total sleep time. Sleep stage 3 is the most important for rest and recovery. During this stage, there is no muscle activity and the brain emits slow brain waves, called delta waves, which are associated with a restful state and deep recovery. A healthy person spends the most time in sleep stage 3 during the first half of the night. During the initial sleep cycle, sleep stage 3 usually lasts 20 to 40 minutes. Throughout the night, the time spent in sleep stage 3 gradually becomes shorter, and more sleep time is spent in sleep stage 4.

[0010] Sleep stage 4 is REM sleep and accounts for about 25% of total sleep time. Sleep stage 4 is the stage where dreams occur. Sleep stage 4 plays a role in enhancing memory, processing emotions, and overall physical recovery.

[0011] During sleep stage 4 (REM), the brain functions in a similar way to when you are awake. Brain waves are active, breathing and heart rate increase, and become more erratic than in other sleep stages, which means the body does not rest during REM sleep. If REM sleep is too long at night, it means that there is a lack of deep sleep in sleep stage 3, which can make people feel tired in the morning.

[0012] REM sleep was previously thought to be the most important sleep stage for learning and memory, but newer research suggests that, aside from its restful and restorative advantages over REM sleep, non-REM sleep (sleep stages 1, 2, and 3) is also more important for learning and memory. Given the surge in activity during REM sleep, this may explain why frail people often experience heart attacks or other illnesses in the early morning, when REM sleep is usually longer.

[0013] Scientific studies have shown that applying slow oscillations to a bed can reduce the time required to start sleep and increase the time in deep sleep states, thereby improving a person's memory. The applicant's patent application WO2018083153 discloses a method for controlling a rocking bed, stopping the rocking motion of the bed if the user leaves the bed or if the user falls asleep (because they have stopped moving).

[0014] However, each person has a different sensitivity to rocking or oscillating motion. If this motion is excessive for a particular person, that person may experience dizziness, or even nausea or vomiting. This is far from ideal for a relatively strong and healthy person, and can pose a real danger to the elderly or more vulnerable.

[0015] It is an object of the present invention to reduce or substantially eliminate the above mentioned problems. Summary of the invention

[0016] According to a first aspect of the present invention, there is provided a method for controlling a sleep enhancement device or system (on a bed or beside a bed) for improving the sleep quality of a sleeper, the method comprising the following steps:

[0017] Based on information from at least one or more sensors, estimating which sleep stage the sleeper is in, the sleep stage selected from the group consisting of sleep stage 3 (N3), sleep stage 4 (REM), and another sleep stage (optionally, including sleep stage 1 (N1) and / or sleep stage 2 (N2));

[0018] When the sleeper is estimated to be in sleep stage 3 (N3) and / or when the sleeper is estimated to be in sleep stage 4 (REM) (and optionally, when they are estimated to be in sleep stage 1 (N1) and / or sleep stage 2 (N2)), at least one of the following is performed:

[0019] activating a sleep enhancement device or system to begin providing input to act upon the sleeper at the estimated sleep stage;

[0020] deactivating the sleep enhancement device or system to cease providing input to the sleeper at the estimated sleep stage;

[0021] The sleep enhancement device or system is adjusted (or alternatively, maintained) to vary the level of input applied to the sleeper during the estimated sleep stage.

[0022] The present invention has the major advantage of improving overall sleep quality, for example allowing a person to have more restorative sleep. The present invention can extend the duration of sleep in a particular sleep state and / or reduce the duration of sleep in another particular sleep state. That is, the time spent in one sleep stage can be controlled relative to the time spent in another sleep stage.

[0023] In other words, the present invention allows for maximization of time spent in deep sleep and / or REM. In particular, the present invention can allow for a controlled reduction in the proportion of sleep time a person spends in sleep stage 1 (N1) and sleep stage 2 (N2) relative to sleep stage 3 (N3) and / or sleep stage 4 (REM). For example, the present invention can allow a person to reach sleep stage 3 (N3) more quickly than would otherwise be possible.

[0024] The method aims to maintain a deep sleep state of the user of the bed while minimizing adverse effects that may occur due to the oscillatory movement of the bed. That is, the movement of the bed is tailored to the person and / or controlled (preferably automatically) in a manner that depends on the sleep stage the person is estimated to be in.

[0025] Preferably, the instructions or decisions to start and / or stop and / or adjust the input provided by the device / system are determined by an artificial intelligence (AI) algorithm or machine learning model trained on an appropriate number of data sets. Generally speaking, the greater the number of data sets used for training, the better the training results of the final model. For example, the number of data sets can be approximately, and at least approximately, 10,000.

[0026] Based on the information provided by sensors that sense or regularly / continuously monitor vital parameters of the human body, algorithms or models can accurately estimate the sleep stage a person is in and provide instructions / decisions accordingly.

[0027] The present invention may utilize an AI algorithm that, based on information received from sensors during a person's sleep, may learn patterns in a person's sleep cycle and / or sleep stages (e.g., based on data obtained over several nights of sleep by the person). Based on this information, the AI ​​algorithm may predict and determine when input from a sleep enhancement device / system should be started and / or stopped and / or adjusted to improve sleep quality.

[0028] The present invention is suitable for use by anyone suffering from a sleep disorder and / or any secondary health effects experienced or exacerbated by lack of sleep.

[0029] The present invention is also suitable for people who do not have any sleeping problems / disorders, for example, if they wish to reduce the total sleeping time required. That is, to achieve restful sleep in a shorter time.

[0030] In a preferred embodiment, the time a sleeper spends in sleep stages 1 and 2 should be reduced to a minimum, while quickly entering sleep stage 3 and staying there longer than normal, alternating with sleep stage 4 (REM), which is important for memory regeneration.

[0031] Thus, the method can prioritize initiating and / or maintaining sleep stage 3 and provide input to achieve this goal. By providing input, the method can also prioritize artificially extending and / or artificially ending one or both of sleep stages 3 and 4 to control a person's transition between sleep stages 3 and 4 while minimizing the time spent in sleep stages 1 and 2.

[0032] The estimate of a particular sleep stage present at the time of a given sensor reading can be based on a confidence score / value. That is, the sensor reading can be processed or evaluated to determine the likelihood that it corresponds to sleep stage 3 (N3) or sleep stage 4 (REM), or another one or more sleep stages, and the sleep stage with the corresponding highest confidence score / value is determined as the estimated sleep stage.

[0033] The estimation may be performed periodically or substantially continuously. For example, the frequency with which the estimation is performed may depend on the data sampling frequency of the sensor and / or the frequency set by the user at which the sleep stage estimation should be refreshed or re-estimated.

[0034] The sleep enhancement device or system may be a bed rocking device or system. For example, bed rocking may be implemented using a device / system similar to that proposed in the applicant's patent application WO2018083153 or the applicant's patent application EP4137008.

[0035] The input may be a bed rocking motion or a reciprocating motion or an oscillating motion. Any of the input values ​​discussed below may correspond to a bed rocking motion velocity or a reciprocating motion velocity or an oscillating motion velocity. In this case, the input value may vary between 0 (i.e., stopped or stationary) and a maximum value of 100 (which may correspond to a velocity of about 10 mm per second).

[0036] If the person is estimated to be in either or both of sleep stage 1 (N1) and sleep stage 2 (N2), the method may include operating (or continuing to operate) the sleep enhancement device or system. That is, the method may allow the sleep enhancement device / system to continue at its previous input value or current state, rather than starting / stopping / adjusting the input.

[0037] The input may be a non-zero input during sleep stage 1 and / or sleep stage 2. For example, in the present method, during any estimated sleep stage 1 and / or sleep stage 2, the bed rocking motion or bed velocity may have a non-zero value.

[0038] In other words, the bed may move or rock during sleep stage 1 and / or sleep stage 2. The input value in sleep stage 1 and / or sleep stage 2 may be a predetermined value or a user-set value. That is, for the operation of the sleep enhancement device during sleep stage 1 and / or 2, the input may be a value preset according to user selection, while for the operation of the sleep enhancement device during sleep stage 3 and / or 4, the input value may be set or changed according to the control data generated or already generated by the machine learning model.

[0039] Rocking the bed to a certain speed during sleep stages 1 and / or 2 can help a person reach sleep stage 3 (N3) more quickly, at which time the bed rocking motion can be reduced or stopped.

[0040] The method may include estimating when a person enters or begins sleep stage 3 (N3) and / or sleep stage 4 (REM) based on information from at least one or more sensors. The method may include stopping or adjusting input provided by a sleep enhancement device or system when it is estimated that the person has entered sleep stage 3 (N3) or sleep stage 4 (REM).

[0041] In other words, when the user is in deep sleep, preferably the movement of the bed is reduced (optionally gradually) to a minimum or stops. This can reduce the adverse effects that the sleep enhancement device or system may have, such as the adverse effects of oscillating / rocking motion.

[0042] The method may include estimating whether sleep stage 3 (N3) and / or sleep stage 4 (REM) has elapsed to a predetermined extent. The method may include activating or adjusting (optionally gradually activating / adjusting) an input provided by a sleep enhancement device or system when it is estimated that the relevant sleep stage has elapsed to a relevant predetermined extent.

[0043] This can help prolong sleep stage 3 and / or end sleep stage 4 relative to what would occur during non-enhanced sleep.

[0044] The predetermined range for sleep stage 3 (N3) may correspond to a few minutes (perhaps up to 10 minutes) before the predicted end of sleep stage 3.

[0045] The predicted end may be based on a determined success rate of previous input starts before the end of sleep stage 3. For example, input (re)starts during sleep stage 3 may occur incrementally or progressively earlier, i.e., before the predicted end of sleep stage 3, extending that sleep stage 3 if input starts in the previous sleep stage did not have the desired effect.

[0046] The sleep enhancement device or system may be programmed with a predetermined input value for sleep stage 3 (N3). The input generated or communicated by the sleep enhancement device / system is set to (optionally incremented to) the predetermined input value around the predicted end of sleep stage 3 (N3).

[0047] This allows user defined input values, such as bed rocking speed, to be implemented based on the sensitivity to the input that the user is comfortable with.

[0048] Any predetermined input values ​​may be set by a remote control (which may be provided with the sleep enhancement device / system) and / or an application on an electronic device (eg, a smartphone or tablet) that is programmed to communicate with the sleep enhancement device / system.

[0049] When the sleeper is estimated to be in sleep stage 3 (N3), if the duration of sleep stage 3 (N3) exceeds about 60 minutes or exceeds a predetermined time length, the input may be stopped. The predetermined time length may be a user-set value or may be a clinically set value (e.g., determined by a doctor), which may be greater than or less than 60 minutes.

[0050] This avoids over-extending sleep stage 3 in order to achieve an optimal reinforcement balance with the other sleep stages and can promote alternation with sleep stage 4.

[0051] If the input is stopped at the end of sleep stage 3 (N3) and the sleeper subsequently enters sleep stage 4 (REM), the input may remain (for a period of time) at a zero input value.

[0052] If the predetermined input value is insufficient to extend the first sleep Stage 3 (N3) period, a higher input value, which may be approximately 5% higher than the predetermined value, may be used during one or more subsequent sleep Stage 3 (N3) periods.

[0053] The sleep enhancement device or system may be programmed with a predetermined input start time for sleep stage 3 (N3). When the sleeper is estimated to be in sleep stage 3 (N3), their input begins at the predetermined input start time.

[0054] These features can in many ways help extend sleep stage 3 to a desired longer duration in order to provide a beneficial longer period of deep sleep.

[0055] The predetermined range for sleep stage 4 (REM) corresponds to approximately half of the predicted duration of sleep stage 4.

[0056] A sleep enhancement device or system may be programmed with a predetermined input value for sleep stage 4 (REM) and have its input set to the predetermined input value approximately midway through sleep stage 4 (REM).

[0057] If the predetermined input value is insufficient to end the first REM sleep period, a higher input value, which may be about 5% higher than the predetermined value, is used during one or more subsequent REM sleep periods.

[0058] The sleep enhancement device or system may be programmed with a predetermined input start time for sleep stage 4 (REM). When the sleeper is estimated to be in sleep stage 4 (REM), the input thereof starts at the predetermined input start time.

[0059] These features may variously help controllably end sleep stage 4 earlier than during non-expanded sleep, and may help facilitate the sleeper's return to a subsequent sleep stage 3 earlier in a given sleep period.

[0060] The one or more other sleep stages include sleep stage 1 (N1) and / or sleep stage 2 (N2). When the sleeper is estimated to be in sleep stage 1 (N1) and / or when the sleeper is estimated to be in sleep stage 2 (N2), the method includes at least one of the following:

[0061] activating a sleep enhancement device or system to begin providing input to the sleeper;

[0062] deactivating the sleep enhancement device or system to cease providing input to the sleeper;

[0063] Keep inputs to the level already provided by the sleep enhancement device or system;

[0064] The sleep enhancement device or system is adjusted to vary the level of input applied to the sleeper.

[0065] During sleep, a person may enter different sleep stages at different times. It will be appreciated that the method involves controlling the input (whether starting, stopping, maintaining and / or adjusting the input) at different points in some or all of the sleep stages, which may include controlling the input at the beginning of a particular sleep stage (based on estimates made in the method) and / or controlling the input midway through a particular sleep stage (based on estimates made in the method), and / or controlling the input at a particular sleep stage (based on estimates made in the method).

[0066] It should also be understood that controlling an input during a sleep stage may also include multiple controls of an input during a single sleep stage.

[0067] The sleep enhancement device or system may be programmed with clinical information of the sleeper and control any one or more of the following input parameters based at least in part on the clinical information: input type, input intensity, input start time, input end time, input duration, input mode.

[0068] A doctor or other medical / nursing professional may provide relevant information to the device / system which may affect its operation. For example, if the sleeper is diagnosed with a sleep condition / disorder, this may override or take precedence over what the control data or instructions may contain.

[0069] The device / system may use clinical information in a manner that attempts to avoid an input that exacerbates an existing condition and / or attempts to minimize the likelihood of triggering another condition. For example, if the user has cardiovascular issues and is at risk of having a heart attack during REM sleep, the device / system may reduce the time it takes to stop the bed rocking during sleep stage 4.

[0070] The sleep enhancement device or system may be programmed with a sleep duration value. The sleep duration is measured from the clock time or the time when it is detected that the sleeper has fallen asleep.

[0071] In the event that a given sleep stage 3 (N3) or sleep stage 4 (REM) is predicted to exceed the last sleep stage 3 (N3) or last sleep stage 4 (REM) before the sleep duration value, the sleep enhancement device or system and related inputs are stopped and not restarted when the sleeper enters any subsequent sleep stage before waking up.

[0072] Thus, the sleeper can set the time they want to fall asleep and the sleep enhancement device / system will operate accordingly. Stopping inputs (such as rocking the bed) at the last predicted deep sleep stage and not restarting them at subsequent sleep stages before waking can help avoid waking a person feeling groggy.

[0073] Any suitable sensor may be used. A consumer or medical device (or devices) may incorporate sensors for monitoring vital parameters that provide sleep stage information. Examples of such devices include any or more of the following for provision in a sleep enhancement system.

[0074] A bracelet or watch may be provided that includes sensors for any, some or all of motion, temperature, measuring heart rate / pulse, electrocardiogram (ECG), blood pressure.

[0075] A chest strap may be provided that includes sensors for any, some or all of motion, temperature, measuring heart rate / pulse, electrocardiogram (ECG), heart rate variability.

[0076] A headband may be provided that includes sensors for any, some or all of motion, temperature and EEG (electroencephalogram).

[0077] A pad type sensor may be provided for placement on or under a mattress. The pad type sensor may include sensors for any, some or all of motion, respiration rate, heart rate / pulse, heart rate variability.

[0078] An accelerometer type sensor may be provided for positioning on a mattress to sense motion. The purpose of the sensor is to sense motion of a person on the mattress, which motion may occur when the person on the mattress changes lying position, such as rolling over. For example, an accelerometer type sensor is not intended to sense lateral motion of the mattress caused by a bed rocking device or apparatus.

[0079] All of the above sensors have various advantages and disadvantages due to their positioning relative to a person's body during sleep. Some of them may be uncomfortable during sleep (such as chest straps or headbands). Regardless of which device a person is provided and wears during sleep, the positioning of each device relative to the person's body and the type of device used will affect the accuracy and usefulness of the readings of the person's vital parameters.

[0080] Since there is currently no universal sensor that can properly sense all vital parameters of the body, using a combination of multiple sensors (preferably 3-6 or more different sensors) means that the estimation of sleep stages can be more accurate.

[0081] The method may be a computer implemented method.

[0082] The method may be performed on a bed or seat where a sleeper is located.

[0083] The input may include rocking or reciprocating motion.

[0084] The method may be performed at least in part using control data generated by a machine learning model according to the third aspect of the invention to estimate one or more sleep stage parameters.

[0085] According to a second aspect of the present invention, there is provided a system for improving the sleep quality of a sleeper, the system comprising:

[0086] one or more sleep enhancement devices configured to provide input for acting on a sleeper;

[0087] one or more sensors on or near the sleeper, each of the one or more sensors being selected from the group consisting of a motion sensor, an accelerometer, a temperature sensor, a respiration rate sensor, an electrocardiogram (ECG) sensor, a heart rate sensor, a pulse sensor, a heart rate variability sensor, a blood pressure sensor, such as an arterial pressure sensor, an electroencephalogram (EEG) sensor;

[0088] A processing device is configured to:

[0089] receiving information from the one or more sensors,

[0090] using the received information to generate or obtain an estimate of which sleep stage the sleeper is in, the sleep stage being selected from sleep stage 3 (N3), sleep stage 4 (REM), or another sleep stage (which may be sleep stage 1 (N1) and / or sleep stage 2 (N2)); and

[0091] During the estimated sleep stage, controlling the one or more sleep enhancement devices includes starting, stopping or changing inputs when the sleeper is estimated to be in sleep stage 3 (N3) and / or sleep stage 4 (REM) (optionally, and sleep stage 1 (N1) and / or sleep stage 2 (N2)).

[0092] The beneficial effects are similar to those discussed with respect to the first aspect of the invention.

[0093] The system may be configured to perform the method of the first aspect of the invention.The system may include any one or more of the features presented in relation to the first aspect of the invention.

[0094] According to a third aspect of the present invention, there is provided a method for generating control data for controlling a sleep enhancement device or system to improve sleep quality by means of a machine learning model, the method comprising the following steps:

[0095] Providing or accessing a trained machine learning model to estimate the sleep stage at which sensor readings are collected, including collecting sensor readings during sleep stage (N3) and collecting sensor readings during sleep stage 4 (REM) (optionally, collecting sensor readings during one or both of sleep stage 1 (N1) and sleep stage 2 (N2)), or training such a machine learning model;

[0096] providing a trained machine learning model with a dataset of multiple sensor readings of a sleeper, the datasets being collected or being collected by one or more types of sensors during one or more sleep cycles;

[0097] The sensor readings of the sleeper are processed by the trained machine learning model, and the trained machine learning model estimates that the sensor readings indicate that the sleeper is in sleep stage 3 (N3), sleep stage 4 (REM), or another sleep stage (which may be sleep stage N1 and / or sleep stage 2 (N2));

[0098] Control data is generated, via a trained machine learning model, based on an estimate that sensor readings indicate that the sleeper is in at least one of sleep stage 3 (N3) and sleep stage 4 (REM) (optionally, and sleep stage 1 (N1) and / or sleep stage 2 (N2)), wherein the control data is customized for the sleeper and is used to control the sleep enhancement device or system and the input provided thereby to act on the sleeper to improve their sleep quality.

[0099] This aspect of the invention generates tailored or customized control data for a particular person's sleep pattern or sleep stage to control the sleep enhancement device / system in an optimal manner for that particular person. This can significantly improve their overall sleep quality by controlling the operation of the sleep enhancement device / system in a manner that complements the sleep stage that the person is in at different times during sleep. Other beneficial effects are similar to those discussed in the first aspect of the invention.

[0100] The processing step includes estimating, by the trained machine learning model, that the sensor reading indicates that the sleeper has entered one of sleep stage 3 (N3), sleep stage 4 (REM), or another sleep stage. The control data includes instructions for stopping or adjusting the input provided by the sleep enhancement device or system when it is estimated that the sleeper has entered sleep stage 3 (N3) or sleep stage 4 (REM).

[0101] This means that when the user is in deep sleep, the movement of the bed can be reduced to a minimum or stopped. This can reduce possible adverse effects caused by the sleep enhancement device or system, such as those caused by oscillating / rocking movements.

[0102] The processing step includes estimating, by means of a trained machine learning model, that the sensor readings indicate that a predetermined degree of sleep stage 3 (N3) and / or sleep stage 4 (REM) has elapsed. The control data includes instructions for activating or adjusting an input provided by the sleep enhancement device or system when it is estimated that the relevant sleep stage has elapsed to a relevant predetermined degree.

[0103] For example, the instructions may cause the bed to rock at a predetermined speed or may increase the bed rocking speed. This helps to extend sleep stage 3 and / or end sleep stage 4 relative to what would happen to a person during non-enhanced sleep.

[0104] The one or more other sleep stages include sleep stage 1 (N1) or sleep stage 2 (N2). The generated control data is based on an estimation that the sensor reading indicates that the sleeper is in sleep stage 3 (N3), sleep stage 4 (REM), and any one of sleep stage 1 (N1) and / or sleep stage 2 (N2).

[0105] The control data includes instructions for controlling the rocking or reciprocating motion of the sleep enhancing device or system, including one or more bed oscillation or reciprocation speed values.

[0106] The instructions in the control data may correspond to various features of the first aspect of the present invention.

[0107] Sensor readings may be taken while the sleeper is sleeping on a bed that has a rocking or reciprocating motion.

[0108] The data set includes a plurality of bed vibration or reciprocating velocity values ​​corresponding to each sensor reading when each corresponding sensor reading was acquired.

[0109] This allows data from a rocking bed to be taken into account when determining how a person's sleep stages change during sleep in a rocking bed, as opposed to a traditional non-rocking bed.

[0110] The sensor readings are from any two, three, four, five, six or more of the following types of sensors: motion sensor, body temperature sensor, heart rate or pulse sensor, blood pressure sensor, EEG sensor, respiration rate sensor, electrocardiogram sensor, heart rate variability sensor.

[0111] As mentioned earlier, using data from multiple sensors can more accurately estimate which sleep stage a person is in at different points during sleep.

[0112] If you train a machine learning model from scratch before processing a specific person’s sleep data, this approach might involve any, some, or all of the following:

[0113] receiving a data set (which may have been cleaned and / or normalized and / or windowed) comprising a plurality of sensor readings acquired by one or more types of sensors during one or more sleep periods (optionally selected from the sensors described above or in the previous aspects of the invention, and optionally including one or more polysomnograms), and optionally a plurality of input readings taken at times corresponding to each of the plurality of sensor readings, each input reading corresponding to an input acting on the person at the time of the sensor reading (where the input may be bed rocking velocity);

[0114] receiving a sleep stage tag for each sensor reading, including a sleep stage 3 (N3) tag associated with each sensor reading acquired during sleep stage 3 (N3) of the sleeper's sleep, and a sleep stage 4 (REM) tag associated with each sensor reading acquired during sleep stage 4 (REM) of the sleeper's sleep, and optionally corresponding tags for readings acquired in sleep stage 1 (N1) and / or sleep stage 2 (N2);

[0115] Optionally, if sensor readings from multiple types of sensors are provided, a common timestamp or timestamps of matching sensor readings are used;

[0116] Optionally, receiving information about a range of sensor reading values ​​and a range of values ​​corresponding to respective sleep stages (e.g., via supervised learning);

[0117] Using each of the plurality of sensor readings, the sleep stage label associated with each sensor reading, and optionally each input reading taken at a time corresponding to each of the plurality of sensor readings, and optionally any other features described above, train a machine learning model to at least identify sensor readings corresponding to sleep stage 3 (N3) and sensor readings corresponding to sleep stage 4 (REM) (and optionally sensor readings corresponding to one or both of sleep stage 1 (N1) and sleep stage 2 (N2); which may include training using any, some or all of a clustering module, a case-based reasoning module, a data mining module, an LSTM (long short-term memory) neural network module, each of which is a known or publicly available module;

[0118] Optionally, the training step comprises training a machine learning model to estimate or determine sleep stages in which a given sensor reading is obtained based on sensor reading values ​​from two or more different types of sensors (unsupervised learning).

[0119] It is worth noting that supervised learning occurs by feeding previously captured data from sensors and desired output bed motion control data to the algorithm with the goal of training the model to produce the desired output, whereas unsupervised learning occurs by the algorithm learning from scratch and deciding which is the best solution to control the bed motion given that the sleep characteristics of a particular subject may not fit into previously known patterns or groups.

[0120] According to a fourth aspect of the present invention, there is provided a non-transitory computer-readable medium, comprising program instructions for causing a processing device to execute a method for controlling a sleep enhancement device or system to improve sleep quality, the instructions comprising:

[0121] Instructions for estimating which sleep stage the sleeper is in based on information from at least one or more sensors, the sleep stage selected from sleep stage 3 (N3), sleep stage 4 (REM), and one or more other sleep stages (optionally, sleep stage 1 (N1) and sleep stage 2 (N2)); and

[0122] When the sleeper is estimated to be in sleep stage 3 (N3) and / or when the sleeper is estimated to be in sleep stage 4 (REM) (optionally, sleep stage 1 (N1) and sleep stage 2 (N2)), instructions for performing at least one of the following:

[0123] At the estimated sleep stage, activating a sleep enhancement device or system to begin providing input to the sleeper;

[0124] At the estimated sleep stage, stopping the sleep enhancement device or system to stop providing input to the sleeper;

[0125] During the estimated sleep stage, the sleep enhancement device or system is adjusted to change the level of input to the sleeper.

[0126] The beneficial effects are similar to those of the first and second aspects of the present invention.

[0127] According to a fifth aspect of the present invention, there is provided a non-transitory computer-readable medium, comprising program instructions for causing a processing device to execute a method for generating control data through a machine learning model to control a sleep enhancement device or system, thereby improving sleep quality, the instructions comprising:

[0128] Instructions for providing or accessing a machine learning model trained to estimate sleep stages when collecting sensor readings, including collecting sensor readings during sleep stage (N3) and collecting sensor readings during sleep stage 4 (REM) (optionally collecting sensor readings during sleep stage 1 (N1) and collecting sensor readings during sleep stage 2 (N2)), or training such a machine learning model;

[0129] Instructions for providing a trained machine learning model with a dataset of a plurality of sensor readings of a sleeper, the datasets being or being collected by one or more types of sensors during one or more sleep cycles;

[0130] Instructions for processing sensor readings of a sleeper by a trained machine learning model, and estimating by the trained machine learning model that the sensor readings indicate that the sleeper is in one of sleep stage 3 (N3), sleep stage 4 (REM), or other sleep stages (optionally, sleep stage 1 (N1) and / or sleep stage 2 (N2));

[0131] Instructions for generating control data based on an estimate by a trained machine learning model that a sleeper is in at least one of sleep stage 3 (N3) and sleep stage 4 (REM) (optionally, sleep stage 1 (N1) and / or sleep stage 2 (N2)), the control data being customized for the sleeper for controlling a sleep enhancement device or system and the input provided thereby to act on the sleeper to improve their sleep quality.

[0132] Its beneficial effects are similar to those of the fourth aspect of the present invention.

[0133] The present invention provides a sleep enhancement device or system (such as a bed rocking device, a rocking chair, or a bed motion leg), comprising the non-transitory computer-readable medium described in the fourth or fifth aspect of the present invention.

[0134] Any aspect of the present invention may include any feature or independently selected combination of features presented in relation to any other aspect of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0135] For a better understanding of the invention, and in order to more clearly show how the invention may be put into practice, reference will now be made to the accompanying drawings, which will be presented by way of example only, in which:

[0136] Figure 1 is a perspective view of a person lying on a bed including a system for rocking the bed;

[0137] Figure 2 is a flow chart for processing data associated with a person sleeping in a bed to provide bed motion control data that can be used in a predictive model for how to control the rocking motion of the bed;

[0138] Figure 3 is a flow chart for using bed motion control data for rocking of the bed according to predicted sleep stages. DETAILED DESCRIPTION

[0139] See first Figure 1 , a bed is generally indicated at 10. A person 12 lies on the bed 10. The bed 10 has a mattress, a pillow, a headboard and a base.

[0140] In this embodiment, four bed legs 11 are provided under the four corners of the base. The bed legs 11 are motion-applying legs that are capable of applying a reciprocating lateral rocking motion to the bed 10, so that the bed legs swing back and forth in a horizontal plane during operation. Examples of the types of legs that can achieve this are given in the applicant's patent application WO2018083153 or the applicant's patent application EP4137008, the disclosures of each of which are incorporated herein by reference.

[0141] The legs 11 may be interconnected (e.g. wirelessly) and preferably operate in a synchronized manner so that the legs 11 on the left side of the bed 10 work in opposition to the legs 11 on the right side of the bed 10 when the bed is rocked. During use, when the legs 11 are engaged, the bed 10 may move laterally left and right in a reciprocating manner within the range of displacement provided by the legs.

[0142] The depicted bed arrangement has five different types of sensor devices to sense various parameters associated with the person 12 for estimating which sleep stage they are in at a given time during sleep. The sensor devices in this embodiment include: a watch or bracelet 14, a chest strap 16, a head strap 18, a pad type sensor 20, and an accelerometer 22. It should be understood that the form of each type of device is not limited to the depicted version.

[0143] The watch or bracelet 14 may have any sensor for motion, temperature, measuring heart rate / pulse, electrocardiogram (ECG), and blood pressure. The chest strap 16 may have any sensor for motion, temperature, measuring heart rate / pulse, electrocardiogram (ECG), and heart rate variability (derivable from ECG). The headband 18 may have any sensor for motion, temperature, and EEG (electroencephalogram). The pad-type sensor 20 may have any sensor for motion, breathing rate, heart rate / pulse, and heart rate variability.

[0144] It should be understood that any one, two, three or four or all five sensor types of the sensor device may be used in any embodiment, but in some cases it is preferred to use multiple sensors. It will also be understood that some sensor devices may be provided in a combination device where a given sensor device has sensors that are the same as two of the different sensors shown. Examples of other sensor devices include smartphones and pulse oximeters.

[0145] This type of arrangement can be used with many different people to record sleep data covering a wide range of body parameters during a range of sleep cycles.

[0146] Regarding supervised training, the table below shows a mapping of the information obtained from each type of sensor to each sleep stage (N1, N2, N3, REM). However, ECG and heart rate variability sensors currently require specialized analysis algorithms (optionally using artificial intelligence) to interpret how the sensor readings correspond to sleep stages, so they are not summarized below.

[0147] Table 1 Correspondence between sleep stages and corresponding expected sensor reading values ​​for motion sensor, body temperature sensor and heart rate / pulse sensor

[0148]

[0149]

[0150] Table 2 Correspondence between sleep stages and corresponding expected sensor reading values ​​for blood pressure sensor, EEG sensor and respiratory rate sensor

[0151]

[0152] That is, in this case, the AI ​​algorithm is told to which sleep stage each sensor reading from those sensors in the table belongs.

[0153] It is worth noting that there may be a step where a doctor or intelligent AI software (which is capable of analyzing sleep data, such as from a full night of sleep data) provides or determines sleep stage information or decisions about ECG and / or heart rate variability sensor readings. The ECG and / or heart rate variability data and the corresponding sleep stage for each epoch (30 or 60 seconds) can then be fed to the AI ​​algorithm for supervised training.

[0154] After collecting sensor data from a suitable volume sample of different people, the data can be used as part of training a machine learning model to estimate which sleep stage a sleeper is in based on sensor readings from one or more sensors.

[0155] In the case of supervised learning, the machine learning model can be provided with sensor data and corresponding sleep stage information (e.g., derived from the sensor data or determined by a physician). The sensor data can be replaced or supplemented with polysomnographic data. This is advantageous because the sleep stage information obtained from polysomnography is more accurate, thereby improving the training of the machine learning model.

[0156] In the case of unsupervised learning, a machine learning model can be trained without input of sleep stage information, such that the output of the model / AI algorithm is a determination or estimate of the sleep stage attributed to each sensor reading. However, this is currently not as accurate as sleep stage information determined by the other means described above, so supervised learning of the machine learning model may produce better results.

[0157] Figure 2 A flowchart 100 depicts the general steps required for the process. Flowchart 100 corresponds to the proposed predictive model learning method implementing artificial intelligence and machine learning algorithms. The method generates a predictive model based on input data obtained from sensors (optionally, together with instantaneous bed oscillation velocity information) and produces control data that sets the bed rocking motion velocity value as output.

[0158] When preparing the data collected for this purpose, the instantaneous sleep stage of the analyzed subject in each case can be accurately estimated by using known methods in sleep science (i.e. by a medically trained professional / doctor, or by intelligent artificial intelligence software - preferably using polysomnographic data) in order to derive the sleep stage level starting from the dedicated and specific information provided by each type of sensor.

[0159] It should be understood that artificial intelligence algorithms can achieve:

[0160] Supervised learning, which in this case may be performed by feeding previously captured data from sensors and the desired output control data (i.e. how to control the movement of the bed via the legs 11) to the algorithm with the aim of training the model to produce the desired output; and / or

[0161] Unsupervised learning,With unsupervised learning, the algorithm learns from scratch and determines which is the best solution to control the movement of the bed, given the subject’s specific sleeping characteristics that may not fit into previously known patterns or clusters.

[0162] It should be understood that the quality of the training data set is very important for accurately training AI algorithms. Therefore, as described above, in some preferred embodiments, it is advantageous to obtain / use polysomnography for use in supervised training as well as data from any of the above sensors. Polysomnography captures important parameters of the human body very accurately, and obtaining sleep stage information based on polysomnography (rather than from the other sensors above) can obtain a more accurate training model.

[0163] It should also be understood that some sensor readings / signals, such as EEG and ECG, may be pre-processed and / or transformed to the frequency domain etc. in order to improve the accuracy of the model (once the data is trained). As part of the process, automatic sleep stage classification by analyzing EEG and ECG signals, if available, is helpful.

[0164] In some embodiments, the sleep stage at which the various sensor readings are recorded may be determined automatically by a software program, or manually by a trained professional (eg, a physician).

[0165] Once the sensor readings have been correctly classified according to the sleep stage to which they correspond, the sensor readings and sleep stage information can be fed to an AI algorithm for supervised training.

[0166] However, it is beneficial to first filter, clean, window, downsample and / or normalize the data. The data cleaning and normalization module shown in flow chart 100 translates / adjusts all data sources from various sensors and bed oscillation motion speed into:

[0167] a) a common sampling frequency that is specific to the important parameters of each type of capture; and

[0168] b) Common data ranges, such as normalizing the common ranges specific to each type of sensor.

[0169] For (a), EEG and ECG sensors may have a sampling frequency (ie, the rate at which readings are taken) in the range of 200 Hz to 500 Hz. Sensors for motion, temperature, heart rate / pulse, heart rate variability, respiratory rate, and / or bed oscillation motion rate may have a sampling frequency of approximately 1 Hz.

[0170] For (b), EEG, ECG, and / or motion sensor readings may be in the range of -1.0 to +1.0. Sensor readings for heart rate / pulse, heart rate variability, and / or respiratory rate may be in the range of 0 to 256. Body temperature sensor readings may be in the range of 35 degrees Celsius to 41 degrees Celsius or 95 degrees Fahrenheit to 106 degrees Fahrenheit. Sensor readings for bed oscillation motion velocity may be in the range of 0 to 100 (which corresponds to 0 to approximately 10 mm per second).

[0171] The data cleaning and normalization module can be used to:

[0172] one or more low pass filters for smoothing the sensor data, and / or

[0173] Anomaly prediction algorithms that filter out noisy or erroneous or unwanted data based on known characteristics of previously received data.

[0174] This results in a cleaner dataset, which in turn leads to more accurate model estimates / predictions (once trained).

[0175] The data cleaning and normalization module can also window the input data into a standard set of lengths (e.g., processing frame size). This can correspond to, for example, 30 seconds or 60 seconds, and the window size value can also be set to another value. Given the large amount of data accumulated over time by running AI algorithms, this can reduce the large memory requirements of the system by downsampling data that carries redundant information.

[0176] The next module represents a person (e.g. a supervisor or AI software engineer) who categorizes the data for supervised or unsupervised processing by setting appropriate labels. This is to produce training, test, blind test, and runtime datasets, which are used for creation, testing, validation / verification, and real-time operation of machine learning models / algorithms, respectively. These are all well-known steps in building datasets for machine learning models.

[0177] The control rules of the classification module are:

[0178] Input data type: training, testing, blind test, validation;

[0179] In the case of supervised training, the algorithm is trained with known sleep stage information

[0180] The control flags for the classification module are:

[0181] supervision or no supervision;

[0182] First time user, or not first time user.

[0183] Each module following the classification module is a well-known AI algorithm that can make predictions with a certain confidence level and meets the requirements for training a machine learning model on the sleep dataset: clustering, case-based reasoning, data mining, LSTM (Long Short-Term Memory) neural network.

[0184] The clustering module is able to classify user sleep into groups covering both disturbed / dysfunctional and normal types of sleep behaviors.

[0185] The case-based reasoning module is able to solve new problems by retrieving stored “cases” that describe scenarios similar to those that previously solved the problem, and adapt the solution to the new requirements.

[0186] The data mining module can extract, deduce or abstract rules from large amounts of data.

[0187] The LSTM neural network is able to make long-term and short-term sleep stage predictions.

[0188] The decision engine then needs to select the most relevant information from each of the four AI modules described above. Together with the control rules and flags set by the user, it then generates sleep stage confidence levels / thresholds and flags for when to start, stop or adjust inputs (e.g., bed movement) in sleep stage 3 and / or sleep stage 4, and optionally sleep stage 1 and / or sleep stage 2.

[0189] It should be noted that the input may be a user-selected preset having a non-zero value in sleep stage 1 and / or sleep stage 2. For example, for a sleep enhancement device including a motion imparting device for reciprocating rocking of the bed, a non-zero speed for the reciprocating motion of the bed may be used.

[0190] Typically, the marker or trigger for movement to initiate or "turn on" Sleep Stage 3 will be set a few minutes before the predicted expiration time of Sleep Stage 3. The exact point at which this is set is based on the assumed success rate of previous initiations. That is, if in a previous Sleep Stage 3, the input was set at a time t-5 minutes from the predicted end time t of Sleep Stage 3, but the user actually transitioned out of Sleep Stage 3 earlier than predicted, then in the subsequent Sleep Stage 3, the input will be scheduled to start appropriately earlier to mitigate the possibility of starting too late again.

[0191] Typically, the marker or trigger for movement to initiate or "turn on" sleep stage 4 will be set to about half of the usual time spent in sleep stage 4, i.e., around the halfway point of sleep stage 4s. The duration of REM cycles is different for each sleep cycle and increases throughout the night. For example, the first sleep cycles have shorter REM sleep periods, while later sleep cycles have relatively longer REM sleep periods. This means that the length of time that bed leg movement ceases for sleep stage 4 may need to be adjusted over a few nights so that the AI ​​algorithm can learn the sleeper's specific REM sleep duration for each sleep cycle.

[0192] The control rules of the decision engine are:

[0193] - The total length of sleep time required.

[0194] This can be set by the user via a remote control of the bed leg 11 or via an application that communicates remotely with the bed leg 11 .

[0195] This information is used to shut down movement in bed during the last sleep cycle, thus avoiding the unwanted effect of dizziness upon waking up.

[0196] - Any previously clinically diagnosed sleep condition / disorder that places the user into a specific category.

[0197] Here, AI-generated clustering information is overlaid with clinical diagnoses provided by physicians.

[0198] - Optionally, a predetermined time (set by the user or a medical professional) for switching on the bed oscillation movement before the end of sleep stage 3.

[0199] - Optionally, a predetermined time (set by the user or a medical professional) for switching on the bed oscillation movement after the start of sleep stage 4.

[0200] The control markup of the decision engine is:

[0201] Take a nap during the day, or sleep at night;

[0202] First time user, or not first time user.

[0203] Figure 3 The behavior of the generation module for generating bed motion control data is listed in (flowchart 200) and Table 3. For each sleep stage, there is a target bed oscillation speed, which is derived from the sleep stage confidence of the runtime estimate.

[0204] Table 3 Example bed oscillation / reciprocation speed values ​​and their modulation according to sleep stages

[0205]

[0206]

[0207] After extensively training a machine learning model as shown above, or otherwise having access to a suitably trained model, using some or all of the sensors or sensor devices ( Figure 1 The sleep data set obtained by the training model can be processed or analyzed by the training model. This can be achieved for the first time by the artificial intelligence model estimating the sleep stage of the person 12 while sleeping, or the artificial intelligence model can be provided with data corresponding to one or several nights of sleep of the person 12.

[0208] Based on its training, the AI ​​model can then take sensor readings from the available sensors and assess the likelihood that person 12 is / was in a state corresponding to sleep stage 1 (N1), sleep stage 2 (N2), sleep state 3 (N3), or sleep stage 4 (REM), and can provide a confidence grade / level for each of these. The confidence levels for all four sleep stages (i.e., the probability that a particular sleep stage is the "active" sleep stage at a given time) must always sum to 100%. It should be noted that only one sleep stage can be active at a time, but there may be edge cases when two of the sleep stages are assigned the same large confidence level.

[0209] Based on its training, the AI ​​model may also take sensor readings from available sensors and identify markers or indications in the data that indicate the beginning and end of various sleep stages, in particular: a marker or indication in the data that indicates the beginning of sleep stage 3, a marker or indication in the data that indicates the beginning of sleep stage 4, a marker or indication in the data that indicates the end of sleep stage 3, and a marker or indication in the data that indicates the end of sleep stage 4. This provides the possibility of estimating the length of each sleep stage for the person, taking into account the changes in the length of the sleep stages as the sleep cycle progresses.

[0210] The control data may be generated by a trained machine learning model based on data corresponding to the person 12. The control data includes instructions on when to start input and / or when to stop input and / or when to adjust (increase / decrease / change) the input, which in this case is the speed of movement applied by the bed legs 11.

[0211] In particular, the control data may start / stop / change the input at the beginning of sleep stage 3, at the beginning of sleep stage 4, near the end of sleep stage 3, and / or approximately midway through sleep stage 4, although other points may additionally or alternatively be selected to act on the input. Figure 3 and Table 3, example criteria for start / stop / change inputs are discussed above.

[0212] Thus, the control data prepared may be used to control the input provided by the bed legs 11 each time a person 12 sleeps in the bed 10, so that the bed legs 11 rock the bed 10 in a manner that improves the person's sleep cycles and the relative abundance of sleep stages within those cycles. In particular, the control data is intended to improve the duration of a person's sleep stage 3 (N3) while they sleep, and to relatively controllably alternate between sleep stages 3 (N3) and sleep stages 4 (REM) while the person 12 sleeps.

[0213] It should be understood that in the implementation of the present invention, the algorithm can be run locally on the processor of the rocking bed 10 (or its bed legs 11), or on a server located on the Internet (e.g., on the cloud), which can include two-way data communication and control between the rocking bed 10 and the cloud.

[0214] If a cloud configuration is used, captured real-time data from the sensors can optionally be sent to the cloud along with the bed's instantaneous oscillatory motion parameters (e.g., speed), and in the other direction, real-time commands / instructions with the bed's motion parameters (e.g., the bed's oscillatory motion speed information) can be sent from the cloud to the rocking bed 10.

[0215] The sensors may be connected directly to the shaker (e.g. to a processor) or the sensors may communicate via other methods (e.g. the internet) to a cloud running an AI algorithm.

[0216] A hardware platform that supports algorithms for neural network computing can be used to implement the present invention. The functional blocks of the method can be independently run on a microservice platform.

[0217] It should be understood that the present invention has potential applications for providing better sleep in a variety of different scenarios. This includes scenarios other than sleeping in a bed. For example, implementing the present invention on a seat in a vehicle such as an airplane, train or ship can help improve the sleep quality of passengers.

[0218] The above-described embodiments are provided by way of example only and various changes and modifications will be apparent to those skilled in the art without departing from the scope of the invention as defined in the appended claims.

Claims

1. A method for controlling a sleep enhancement device or system for improving the sleep quality of a sleeper, comprising the following steps: estimating which sleep stage the sleeper is in based on information from at least one or more sensors, the sleep stage being selected from a combination consisting of sleep stage 3 (N3), sleep stage 4 (REM), and one or more other sleep stages; When the sleeper is estimated to be in at least one sleep stage 3 (N3) and / or when the sleeper is estimated to be in sleep stage 4 (REM), at least one of the following is performed: activating the sleep enhancement device or system to begin providing input to the sleeper; deactivating the sleep enhancement device or system to cease providing input to the sleeper; The sleep enhancement device or system is adjusted to vary the level of input applied to the sleeper.

2. A method for controlling a sleep enhancement device or system as claimed in claim 1, characterized in that The following steps are involved: estimating when the sleeper enters sleep stage 3 (N3) and / or sleep stage 4 (REM) based on information from at least one or more sensors; When the sleeper is estimated to have entered sleep stage 3 (N3) or sleep stage 4 (REM), the input provided by the sleep enhancement device or system is stopped or adjusted.

3. A method for controlling a sleep enhancement device or system as claimed in claim 1 or 2, characterized in that The following steps are involved: estimating whether a predetermined degree of sleep stage 3 (N3) and / or sleep stage 4 (REM) has elapsed; When the relevant sleep stage is estimated to have passed the relevant said predetermined degree, the input provided by the sleep enhancement device or system is initiated or adjusted.

4. A method for controlling a sleep enhancement device or system according to claim 3, characterized in that: The predetermined level of sleep stage 3 (N3) corresponds to a maximum of approximately 10 minutes before the predicted end of sleep stage 3.

5. A method for controlling a sleep enhancement device or system according to claim 4, characterized in that: The sleep enhancement device or system may be programmed with a predetermined input value for sleep stage 3 (N3) and set its input to the predetermined input value near the predicted end of sleep stage 3 (N3).

6. A method for controlling a sleep enhancement device or system according to claim 4 or 5, characterized in that: In a case where the sleeper is estimated to be in sleep stage 3 (N3), if the duration of sleep stage 3 (N3) exceeds about 60 minutes or exceeds a predetermined time length, the input is stopped.

7. A method for controlling a sleep enhancement device or system as claimed in claims 3 to 6, characterized in that: The predetermined degree of sleep stage 4 (REM) corresponds to approximately half of the predicted duration of sleep stage 4 (REM).

8. A method for controlling a sleep enhancement device or system according to claim 7, characterized in that: The sleep enhancement device or system may be programmed with a predetermined input value for sleep stage 4 (REM) and have its input set to the predetermined input value approximately midway through sleep stage 4 (REM).

9. A method for controlling a sleep enhancement device or system according to any one of claims 1 to 8, characterized in that Includes any or both of the following: When dependent on claim 5, if the predetermined input value is insufficient to extend the first sleep Stage 3 (N3) period, using a higher input value during one or more subsequent sleep Stage 3 (N3) periods; as well as When dependent on claim 8, if the predetermined input value is insufficient to end a first sleep Stage 4 (REM) period, a higher input value is used during one or more subsequent sleep Stage 4 (REM) periods.

10. A method for controlling a sleep enhancement device or system according to any one of claims 1 to 9, characterized in that Includes any or both of the following: The sleep enhancement device or system may be programmed with a predetermined input start time for sleep stage 3 (N3), and when the sleeper is estimated to be in sleep stage 3 (N3), its input begins at the predetermined input start time; as well as The sleep enhancement device or system may be programmed with a predetermined input start time for sleep stage 4 (REM), and when the sleeper is estimated to be in sleep stage 4 (REM), its input starts at the predetermined input start time.

11. A method for controlling a sleep enhancement device or system according to any one of claims 1 to 10, characterized in that The one or more other sleep stages include sleep stage 1 (N1) and / or sleep stage 2 (N2), and When the sleeper is estimated to be in sleep stage 1 (N1) and / or when the sleeper is estimated to be in sleep stage 2 (N2), at least one of the following is performed: activating a sleep enhancement device or system to begin providing input to the sleeper; deactivating the sleep enhancement device or system to cease providing input to the sleeper; Keep inputs to the level already provided by the sleep enhancement device or system; The sleep enhancement device or system is adjusted to vary the level of input applied to the sleeper.

12. A method for controlling a sleep enhancement device or system according to any one of claims 1 to 11, characterized in that The sleep enhancement device or system may be programmed with clinical information of the sleeper and control any one or more of the following input parameters based at least in part on the clinical information: input type, input intensity, input start time, input end time, input duration, input mode.

13. A method for controlling a sleep enhancement device or system according to any one of claims 1 to 12, characterized in that: The sleep enhancement device or system may be programmed with a sleep duration value measured from a clock time or a time when the sleeper is detected to have fallen asleep, and In the event that a given sleep stage 3 (N3) or sleep stage 4 (REM) is predicted to exceed the last sleep stage 3 (N3) or the last sleep stage 4 (N3) before the sleep duration value, the sleep enhancement device or system and related inputs are stopped and not restarted when the sleeper enters any subsequent sleep stage before waking up.

14. A method for controlling a sleep enhancement device or system according to any one of claims 1 to 13, characterized in that The method is implemented by a computer on a bed or a chair where the sleeper is located, and the input includes rocking or reciprocating motion.

15. A method for controlling a sleep enhancement device or system according to any one of claims 1 to 14, characterized in that The method is performed at least in part using control data generated by a machine learning model through the method of any of claims 18-24.

16. A system for improving the sleep quality of a sleeper, comprising: one or more sleep enhancement devices configured to provide input for acting on a sleeper; one or more sensors on or near the sleeper, each of the one or more sensors being selected from the group consisting of a motion sensor, an accelerometer, a temperature sensor, a respiration rate sensor, an electrocardiogram (ECG) sensor, a heart rate sensor, a pulse sensor, a heart rate variability sensor, a blood pressure sensor, such as an arterial pressure sensor, an electroencephalogram (EEG) sensor; A processing device is configured to: receiving information from the one or more sensors, using the received information to generate or obtain an estimate of which sleep stage the sleeper is in, the sleep stage selected from sleep stage 3 (N3), sleep stage 4 (REM), and one or more other sleep stages; as well as Controlling the one or more sleep enhancement devices includes starting, stopping or changing input provided thereby when the sleeper is estimated to be in sleep stage 3 (N3) and / or sleep stage 4 (REM).

17. A system for improving the sleep quality of a sleeper as claimed in claim 16, characterized in that Used to perform the method according to any one of claims 1 to 15.

18. A method for generating control data for controlling a sleep enhancement device or system to improve sleep quality by a machine learning model, characterized in that: The method comprises the following steps: Providing or accessing a trained machine learning model to estimate the sleep stage during which sensor readings are collected, including collecting sensor readings during sleep stage (N3) and collecting sensor readings during sleep stage 4 (REM), or training such a machine learning model; providing a trained machine learning model with a dataset of multiple sensor readings of a sleeper, the datasets being collected or being collected by one or more types of sensors during one or more sleep cycles; processing sensor readings of the sleeper by the trained machine learning model, and estimating by the trained machine learning model that the sensor readings indicate that the sleeper is in one of sleep stage 3 (N3), sleep stage 4 (REM), and one or more other sleep stages; Control data is generated, via a trained machine learning model, based on an estimate that sensor readings indicate that the sleeper is in at least one of sleep stage 3 (N3) and sleep stage 4 (REM), wherein the control data is customized for the sleeper for controlling a sleep enhancement device or system and the input provided thereby to act on the sleeper to improve their sleep quality.

19. The method of claim 18, wherein: The processing step includes estimating, by the trained machine learning model, that the sensor reading indicates that the sleeper has entered one of sleep stage 3 (N3), sleep stage 4 (REM), or another sleep stage; The control data includes instructions to stop or adjust input provided by the sleep enhancement device or system when the sleeper is estimated to have entered sleep stage 3 (N3) or sleep stage 4 (REM).

20. The method according to claim 18 or 19, characterized in that: The processing step includes estimating, by the trained machine learning model, that the sensor reading indicates that a predetermined degree of sleep stage 3 (N3) and / or sleep stage 4 (REM) has elapsed; The control data includes instructions for activating or adjusting an input provided by the sleep enhancement device or system when the associated sleep stage is estimated to have passed an associated predetermined degree.

21. The method according to claims 18-20, characterized in that One or more other sleep stages include sleep stage 1 (N1) or sleep stage 2 (N2), and the generated control data is based on an estimate that the sensor reading indicates that the sleeper is in sleep stage 3 (N3), sleep stage 4 (REM) and any one of sleep stage 1 (N1) and / or sleep stage 2 (N2).

22. The method according to claims 18-21, characterized in that The control data includes instructions for controlling the rocking or reciprocating motion of the sleep enhancing device or system, including one or more bed oscillation or reciprocation speed values.

23. The method of claim 22, wherein The sensor readings have been or are being collected while the sleeper is asleep on the oscillating or reciprocating bed, and the data set includes a plurality of bed vibration or reciprocating velocity values ​​corresponding to each sensor reading when each corresponding sensor reading was collected.

24. The method according to any one of claims 18 to 23, characterized in that The sensor readings are from any two, three, four, five, six or more of the following types of sensors: motion sensor, body temperature sensor, heart rate or pulse sensor, blood pressure sensor, EEG sensor, respiratory rate sensor, electrocardiogram sensor, heart rate variability sensor.

25. A non-transitory computer readable medium containing program instructions for causing a processing device to execute a method of controlling a sleep enhancement device or system to improve sleep quality, the instructions comprising: Instructions for estimating which sleep stage the sleeper is in based on information from at least one or more sensors, the sleep stage selected from the group consisting of sleep stage 3 (N3), sleep stage 4 (REM), and one or more other sleep stages; as well as When the sleeper is estimated to be in sleep stage 3 (N3) and / or when the sleeper is estimated to be in sleep stage 4 (REM), instructions for performing at least one of the following: activating a sleep enhancement device or system to begin providing input to the sleeper; deactivating the sleep enhancement device or system to cease providing input to the sleeper; The sleep enhancement device or system is adjusted to vary the level of input applied to the sleeper.

26. A non-transitory computer-readable medium comprising program instructions for causing a processing device to execute a method for generating control data through a machine learning model to control a sleep enhancement device or system, thereby improving sleep quality, the instructions comprising: Instructions for providing or accessing a machine learning model trained to estimate sleep stages for collecting sensor readings, including collecting sensor readings during sleep stage (N3) and collecting sensor readings during sleep stage 4 (REM), or training such a machine learning model; Instructions for providing a trained machine learning model with a dataset of a plurality of sensor readings of a sleeper, the datasets being or being collected by one or more types of sensors during one or more sleep cycles; Instructions for processing sensor readings of a sleeper by the trained machine learning model and estimating by the trained machine learning model that the sensor readings indicate that the sleeper is in one of sleep stage 3 (N3), sleep stage 4 (REM), and one or more other sleep stages; Instructions for generating control data based on an estimate by a trained machine learning model that a sleeper is in at least one of sleep stage 3 (N3) and sleep stage 4 (REM) based on sensor readings indicating that the sleeper is in the sleep stage 3 (N3) and the sleep stage 4 (REM), wherein the control data is customized for the sleeper and used to control the sleep enhancement device or system and the input provided thereby to act on the sleeper to improve their sleep quality.

27. A sleep enhancement device or system comprising the non-transitory computer readable medium of claim 25 or 26.

Images