Infant soothing method, apparatus, device and storage medium

Abstract

The application discloses a baby lulling method, device, equipment and storage medium, the baby lulling method comprises: obtaining multi-dimensional information related to baby sleep on a baby bed; based on a preset information fusion model, the multi-dimensional information is subjected to multi-modal fusion processing to obtain target baby state information; based on a preset decision model, the target baby state information is analyzed and processed to obtain a target lulling decision scheme, and the baby bed is controlled to execute a lulling operation for the baby on the baby bed according to the target lulling decision scheme. The application adopts a model training mode of reinforcement learning, and optimizes the decision model algorithm according to the baby state change score before and after the execution of the predicted lulling decision scheme as instant feedback, so that the decision model can be trained independently in the environment without historical lulling data, there is no other condition limitation, and the application is suitable for lulling of all babies, so that the flexibility of the baby bed control is improved.

Description

Technical Field

[0001] The present application relates to the field of computer processing technology, and in particular to a method, device, equipment and storage medium for coaxing an infant to sleep. Background Art

[0002] At present, many families purchase electric rocking beds or electric rocking chairs to help their babies fall asleep instead of parents' hands. However, most electric rocking beds or electric rocking chairs cannot be controlled and require manual control of the switch, and the rocking frequency and amplitude are relatively fixed.

[0003] In related technologies, a technical solution is used to control the crib by matching the current time and current infant status information with historical coaxing data to determine the coaxing operation to be performed by the crib, and then controlling the crib to perform the corresponding coaxing operation. However, this solution requires the collection of historical coaxing data in advance to achieve crib control. For infants without coaxing data, control cannot be achieved. Moreover, if there is little historical coaxing data recorded, the control of the crib will be fixed and repetitive, which limits the control of the crib and results in poor flexibility. Summary of the Invention

[0004] The main purpose of this application is to provide a method, device, equipment and storage medium for coaxing a baby to sleep, aiming to solve the technical problem in the prior art that there are limitations on the control of the baby crib, resulting in poor flexibility.

[0005] To achieve the above objectives, the present application provides a method for coaxing an infant to sleep, the method comprising:

[0006] Obtain multidimensional information related to infant sleep in the crib;

[0007] Based on a preset information fusion model, performing multimodal fusion processing on the multi-dimensional information to obtain target infant status information;

[0008] Analyzing and processing the target infant status information based on a preset decision model to obtain a target sleep-coaxing decision plan, and controlling the crib to perform a sleep-coaxing operation for the infant in the crib according to the target sleep-coaxing decision plan;

[0009] Among them, the decision model is obtained according to the following steps: inputting the baby status information sample into the preset first model to be trained to obtain a predicted coaxing to sleep decision plan; secondly, after controlling the baby crib to execute the predicted coaxing to sleep decision plan, obtaining the updated baby status information sample; then determining the state change score value based on the baby status information sample and the updated baby status information sample; finally, judging whether the state change score value meets the preset first loss convergence condition. If the state change score value does not meet the preset first loss convergence condition, returning to the step of inputting the baby status information sample into the preset first model to be trained to obtain a predicted coaxing to sleep decision plan, until the state change score value meets the preset first loss convergence condition, obtaining a decision model that meets the accuracy condition; the baby status information sample is obtained by performing multimodal fusion processing on the baby coaxing to sleep information sample based on a preset information fusion model, and the first model to be trained is a reinforcement learning model.

[0010] Optionally, the multi-dimensional information includes infant body information, time information, and environmental information related to infant sleep.

[0011] Optionally, before the step of inputting the infant status information sample into a preset first to-be-trained model to obtain a predicted sleep-coaxing decision plan, the method includes:

[0012] performing random information erasure processing on the baby coaxing to sleep information under the baby status information sample to obtain an erased baby coaxing to sleep information sample;

[0013] The erased baby coaxing to sleep information samples include some baby coaxing to sleep information samples of the baby status information samples, or the erased baby coaxing to sleep information samples include all baby coaxing to sleep information samples of the baby status information samples.

[0014] Optionally, after the step of determining a status change score value based on the infant status information sample and the updated infant status information sample, the method includes:

[0015] After controlling the crib to execute the predicted coaxing-to-sleep decision plan, obtaining a coaxing-to-sleep result for the baby;

[0016] The step of determining whether the state change score value satisfies a preset first loss convergence condition, and if the state change score value does not satisfy the preset first loss convergence condition, returning to the step of inputting the infant state information sample into a preset first to-be-trained model to obtain a prediction sleep decision plan, until the state change score value satisfies the preset first loss convergence condition and a decision model that satisfies the accuracy condition is obtained, includes:

[0017] Determining whether the state change score value satisfies a preset first loss convergence condition, and determining whether the baby sleeping result satisfies a preset second loss convergence condition;

[0018] If the state change score value does not meet the preset first loss convergence condition or the baby sleeping result does not meet the preset second loss convergence condition, return to the step of inputting the baby state information sample into the preset first to-be-trained model to obtain the step of predicting the sleeping decision plan, until the state change score value meets the preset first loss convergence condition and the baby sleeping result meets the preset second loss convergence condition, and a decision model that meets the accuracy conditions is obtained.

[0019] Optionally, before the step of obtaining multi-dimensional information related to infant sleep, the method includes:

[0020] Get sample baby sleep information;

[0021] Based on the infant sleep coaxing information sample, the preset second to-be-trained model and the preset first to-be-trained model are jointly trained using a reinforcement learning method to obtain an information fusion model and a decision model that meet the accuracy conditions;

[0022] The second model to be trained is the initial training model of the information fusion model, and the first model to be trained is the initial training model of the decision model.

[0023] Optionally, the step of jointly training the preset second model to be trained and the preset first model to be trained using reinforcement learning based on the infant coaxing-to-sleep information sample to obtain an information fusion model and a decision model that meet accuracy requirements includes:

[0024] Inputting the infant sleep coaxing information sample into a preset second to-be-trained model to obtain an infant status information sample;

[0025] Inputting the infant status information sample into a preset first model to be trained to obtain a prediction and decision scheme for coaxing the infant to sleep, wherein the first model to be trained and the second model to be trained are both reinforcement learning models;

[0026] After controlling the crib to execute the predictive coaxing-to-sleep decision plan, obtaining an updated infant status information sample, and determining a status change score value based on the infant status information sample and the updated infant status information sample;

[0027] Determine whether the state change score value meets the preset first loss convergence condition. If the state change score value does not meet the preset loss convergence condition, return to the step of inputting the baby sleep information sample into the preset second model to be trained to obtain the baby state information sample, until the state change score value meets the preset first loss convergence condition, and obtain an information fusion model and a decision model that meet the accuracy conditions.

[0028] The present application also provides a baby sleeping device, the baby sleeping device comprising:

[0029] an acquisition module, for acquiring multi-dimensional information related to the baby's sleep in the crib;

[0030] a fusion module, configured to perform multimodal fusion processing on the multi-dimensional information based on a preset information fusion model to obtain target infant status information;

[0031] an execution module, configured to analyze and process the target infant status information based on a preset decision model to obtain a target sleep-coaxing decision plan, and control the crib to perform a sleep-coaxing operation on the infant in the crib according to the target sleep-coaxing decision plan;

[0032] Among them, the decision model is obtained according to the following steps: inputting the baby status information sample into the preset first model to be trained to obtain a predicted coaxing to sleep decision plan; secondly, after controlling the baby crib to execute the predicted coaxing to sleep decision plan, obtaining the updated baby status information sample; then determining the state change score value based on the baby status information sample and the updated baby status information sample; finally, judging whether the state change score value meets the preset first loss convergence condition. If the state change score value does not meet the preset first loss convergence condition, returning to the step of inputting the baby status information sample into the preset first model to be trained to obtain a predicted coaxing to sleep decision plan, until the state change score value meets the preset first loss convergence condition, obtaining a decision model that meets the accuracy condition; the baby status information sample is obtained by performing multimodal fusion processing on the baby coaxing to sleep information sample based on a preset information fusion model, and the first model to be trained is a reinforcement learning model.

[0033] And / or, the baby coaxing sleep device further includes: an information erasing module, configured to perform random information erasing processing on the baby coaxing sleep information under the baby status information sample to obtain an erased baby coaxing sleep information sample; wherein the erased baby coaxing sleep information sample includes a portion of the baby status information sample, or the erased baby coaxing sleep information sample includes all of the baby coaxing sleep information samples of the baby status information sample;

[0034] And / or, the baby coaxing sleep device further includes: a coaxing sleep result acquisition module, used to obtain the baby coaxing sleep result after controlling the baby crib to execute the predicted coaxing sleep decision plan; a judgment module, used to judge whether the state change score value meets the preset first loss convergence condition, and judge whether the baby coaxing sleep result meets the preset second loss convergence condition; a second iterative training module, used to return to the step of inputting the baby state information sample into the preset first to-be-trained model to obtain the predicted coaxing sleep decision plan if the state change score value does not meet the preset first loss convergence condition or the baby coaxing sleep result does not meet the preset second loss convergence condition, until the state change score value meets the preset first loss convergence condition and the baby coaxing sleep result meets the preset second loss convergence condition, thereby obtaining a decision model that meets the accuracy condition;

[0035] And / or, the infant coaxing device to sleep further includes: a joint training module for jointly training a preset second model to be trained and a preset first model to be trained using a reinforcement learning method based on the infant coaxing information sample to obtain an information fusion model and a decision model that meet accuracy requirements; wherein the second model to be trained is an initial training model for the information fusion model, and the first model to be trained is an initial training model for the decision model;

[0036] And / or, the joint training module includes: a second state information sample determination module, used to input the baby sleep coaxing information sample into a preset second model to be trained to obtain a baby state information sample; a second scheme prediction module, used to input the baby state information sample into a preset first model to be trained to obtain a predicted sleep coaxing decision scheme, wherein the first model to be trained and the second model to be trained are both reinforcement learning models; a second change scoring module, used to obtain an updated baby state information sample after controlling the baby crib to execute the predicted sleep coaxing decision scheme, and determine a state change score value based on the baby state information sample and the updated baby state information sample; a third iterative training module, used to determine whether the state change score value meets a preset first loss convergence condition. If the state change score value does not meet the preset loss convergence condition, return to the step of inputting the baby sleep coaxing information sample into the preset second model to be trained to obtain the baby state information sample, until the state change score value meets the preset first loss convergence condition, and obtain an information fusion model and decision model that meet the accuracy condition.

[0037] The present application also provides a baby sleeping device, the baby sleeping device comprising: a memory, a processor, and a program stored in the memory for implementing the baby sleeping method.

[0038] The memory is used to store a program for implementing a method for coaxing a baby to sleep;

[0039] The processor is used to execute a program for implementing the method for coaxing a baby to sleep, so as to implement the steps of the method for coaxing a baby to sleep.

[0040] The present application also provides a storage medium, on which is stored a program for implementing a method for coaxing a baby to sleep. The program for implementing the method for coaxing a baby to sleep is executed by a processor to implement the steps of the method for coaxing a baby to sleep.

[0041] The present application provides a method, device, equipment and storage medium for coaxing a baby to sleep. Compared with the related art, it is necessary to collect historical coaxing data of the baby in advance to control the baby crib. For the baby without collected coaxing data, the corresponding control cannot be achieved, and if there are few data records of historical coaxing data, the control of the baby crib will be fixed and repetitive, which limits the control of the baby crib by this scheme, resulting in poor flexibility. In this application, multi-dimensional information related to the baby's sleep in the crib is obtained; based on a preset information fusion model, the multi-dimensional information is multimodally fused to obtain target baby status information; based on a preset decision model, the target baby status information is analyzed and processed to obtain a target coaxing decision plan, and the crib is controlled to perform the coaxing operation for the baby in the crib according to the target coaxing decision plan; wherein, the decision model is obtained according to the following steps: The baby state information sample is input into the preset first model to be trained to obtain a predicted coaxing to sleep decision plan; secondly, after controlling the crib to execute the predicted coaxing to sleep decision plan, an updated baby state information sample is obtained; then based on the baby state information sample and the updated baby state information sample, a state change score value is determined; finally, it is judged whether the state change score value meets the preset first loss convergence condition. If the state change score value does not meet the preset first loss convergence condition, the step of returning to inputting the baby state information sample into the preset first model to be trained to obtain a predicted coaxing to sleep decision plan is performed until the state change score value meets the preset first loss convergence condition, and a decision model that meets the accuracy condition is obtained; the baby state information sample is obtained by performing multimodal fusion processing on the baby coaxing to sleep information sample based on a preset information fusion model, and the first model to be trained is a reinforcement learning model. That is, in this application, a reinforcement learning model training method is adopted, and the decision model algorithm is optimized based on the infant state change score before and after the execution of the predicted coaxing to sleep decision plan as instant feedback, so that the decision model can be trained by autonomously exploring the environment without the need for historical coaxing to sleep data. There are no other conditions or restrictions, and it is suitable for coaxing all infants to sleep, thereby improving the flexibility of crib control. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The accompanying drawings herein are incorporated into and constitute a part of the specification, illustrate embodiments consistent with the present application, and together with the specification, are used to explain the principles of the present application. In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following is a brief introduction to the drawings required for the embodiments or the description of the prior art. Obviously, for those of ordinary skill in the art, other drawings can be obtained based on these drawings without inventive work.

[0043] Figure 1 This is a schematic diagram of the device structure of the hardware operating environment involved in the embodiment of the present application;

[0044] Figure 2 This is a flow chart of the first embodiment of the method for coaxing a baby to sleep;

[0045] Figure 3 This is a schematic diagram of the modules of the baby sleeping device of this application;

[0046] Figure 4 This is a flowchart of a second embodiment of the method for coaxing a baby to sleep according to the present application;

[0047] Figure 5 This is a flowchart of the third embodiment of the method for coaxing a baby to sleep of the present application.

[0048] The realization of the objectives, functional features and advantages of this application will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0049] It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application.

[0050] like Figure 1 As shown, Figure 1 It is a schematic diagram of the terminal structure of the hardware operating environment involved in the embodiment of the present application.

[0051] The terminal in the embodiment of the present application can be a PC, or it can be a smart phone, tablet computer, e-book reader, MP3 (Moving Picture Experts Group Audio Layer III, Moving Picture Experts Compression Standard Audio Layer 3) player, MP4 (Moving Picture Experts Group Audio Layer IV, Moving Picture Experts Compression Standard Audio Layer 4) player, portable computer and other portable terminal devices with display function.

[0052] like Figure 1As shown, the terminal may include: a processor 1001, such as a CPU, a network interface 1004, a user interface 1003, a memory 1005, and a communication bus 1002. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display), an input unit such as a keyboard (Keyboard), and the user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory 1005 may be a high-speed RAM memory or a stable memory (non-volatile memory), such as a disk memory. The memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0053] Optionally, the terminal may also include a camera, an RF (Radio Frequency) circuit, a sensor, an audio circuit, a WiFi module, and the like. Among them, sensors include light sensors, motion sensors, and other sensors. Specifically, the light sensor may include an ambient light sensor and a proximity sensor, wherein the ambient light sensor may adjust the brightness of the display screen according to the brightness of the ambient light, and the proximity sensor may turn off the display screen and / or backlight when the mobile terminal is moved to the ear. As a type of motion sensor, the gravity acceleration sensor can detect the magnitude of acceleration in all directions (generally three axes), and can detect the magnitude and direction of gravity when stationary. It can be used for applications that identify the posture of the mobile terminal (such as horizontal and vertical screen switching, related games, magnetometer posture calibration), vibration recognition related functions (such as pedometer, tapping), etc.; of course, the mobile terminal can also be configured with other sensors such as gyroscopes, barometers, hygrometers, thermometers, infrared sensors, etc., which will not be repeated here.

[0054] Those skilled in the art will understand that Figure 1 The terminal structure shown in the figure does not constitute a limitation to the terminal, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0055] like Figure 1 As shown, the memory 1005 as a computer storage medium may include an operating device, a network communication module, a user interface module and a baby sleeping program.

[0056] exist Figure 1 In the terminal shown, the network interface 1004 is mainly used to connect to the background server and communicate data with the background server; the user interface 1003 is mainly used to connect to the client (user end) and communicate data with the client; and the processor 1001 can be used to call the baby sleep program stored in the memory 1005.

[0057] Reference Figure 2 The embodiment of the present application provides a method for coaxing an infant to sleep, the method comprising:

[0058] Step S100, obtaining multi-dimensional information related to the baby sleeping in the crib;

[0059] Step S200, performing multimodal fusion processing on the multi-dimensional information based on a preset information fusion model to obtain target infant status information;

[0060] Step S300: Analyzing and processing the target baby status information based on a preset decision model to obtain a target sleep-coaxing decision plan, and controlling the crib to perform a sleep-coaxing operation for the baby in the crib according to the target sleep-coaxing decision plan;

[0061] The decision model is obtained by iteratively training a preset first model to be trained using a reinforcement learning method.

[0062] In this embodiment, the application scenarios are:

[0063] As an example, a scenario of coaxing a baby to sleep can be where the baby is coaxed to sleep by the electric crib before the baby lies in the crib and is about to fall asleep. Related art techniques involve matching the current time and current baby state information with historical coaxing data to determine the coaxing operation to be performed by the crib, and then controlling the crib to perform the corresponding coaxing operation to achieve control of the crib. However, this solution requires collecting the baby's historical coaxing data in advance before achieving control of the crib. For babies for whom coaxing data has not been collected, control cannot be achieved. Furthermore, if there are few historical coaxing data records, the control of the crib will be fixed and repetitive, which limits the control of the crib and results in poor flexibility. To address this scenario, the baby coaxing method of this embodiment uses a reinforcement learning model training method and optimizes the decision model algorithm based on the baby's state change score before and after executing the predicted coaxing decision plan as immediate feedback. This allows the decision model to be trained autonomously in the environment without the need for historical coaxing data, without any other restrictions, and is applicable to all babies, thereby improving the flexibility of crib control.

[0064] As an example, the application scenario of coaxing a baby to sleep is not only the above-mentioned application scenario of coaxing a baby to sleep through an electric crib, but also includes various scenarios of coaxing a baby to sleep, which are not specifically limited here.

[0065] This embodiment aims to improve the flexibility of controlling the baby crib.

[0066] In this embodiment, the method for coaxing a baby to sleep is applied to a device for coaxing a baby to sleep.

[0067] The specific steps are as follows:

[0068] Step S100, obtaining multi-dimensional information related to the baby sleeping in the crib;

[0069] As an example, when a baby is lying in an electric crib and is ready to fall asleep, and the parent turns on the baby coaxing function of the crib, the device obtains multi-dimensional information related to the baby's sleep in the crib, wherein the multi-dimensional information related to the baby's sleep includes at least two dimensions of information. Specifically, the multi-dimensional information related to the baby's sleep includes baby's body information, time information, and environmental information related to the baby's sleep. The baby's body information includes but is not limited to information on whether the eyes are open or closed, information on whether the mouth is open or closed, information on the direction of the eye gaze, body posture information, and sound information. Time information refers to the current time. The environmental information related to the baby's sleep includes but is not limited to temperature information, humidity information, lighting information, etc. around the baby's environment. More specifically, the baby's body information (the baby's facial information, limb information and voice information, etc.) is strongly correlated with the baby's sleeping state; time information is weakly correlated with the baby's sleeping state. The weak correlation between time information and sleeping state can be obtained by statistical data on whether the baby is sleepy or sleeping at different time periods every day. More specifically, since babies of different days have different sleeping times every day, but babies of the same day have roughly the same sleeping preferences, it is possible to statistically establish a relationship function based on babies of different days; the environmental information related to the baby's sleep is weakly correlated with the current baby's sleeping state, that is, the environmental information related to the baby's sleep can be used to assist in judging the baby's sleeping state, and cannot determine its discriminant role, that is, the correlation coefficient is low, but it can still be used.

[0070] As an example, the device may obtain multi-dimensional information related to the sleep of a baby in a crib by collecting corresponding information based on relevant sensors (such as cameras, sound sensors, temperature sensors, humidity sensors, clocks, etc.) set on (around) the crib. For example, the device may collect real-time images of the baby and images of the baby's surrounding environment based on a preset camera, and perform image analysis to obtain information on the baby's eye opening and closing, mouth opening and closing, eye gaze direction, human body posture, and ambient lighting. It may also be obtained by receiving multi-dimensional information related to the sleep of a baby in a crib uploaded by a user.

[0071] It should be noted that the device obtains multi-dimensional information related to the baby's sleep in the crib only after obtaining authorization from the user (or parent), and the user has been informed that the collected multi-dimensional information related to the baby's sleep in the crib will be used to control the crib to coax the baby to sleep, and the collected multi-dimensional information related to the baby's sleep in the crib will not be used in other scenarios unrelated to this application.

[0072] Step S200, performing multimodal fusion processing on the multi-dimensional information based on a preset information fusion model to obtain target infant status information;

[0073] As an example, the device performs multimodal fusion processing on the multidimensional information based on a preset information fusion model to obtain the target baby state information, wherein the information fusion model can be trained using a deep neural network learning method of reinforcement learning, or it can be trained using a supervised and unsupervised deep neural network learning method. The training of the information fusion model can be based on the baby coaxing to sleep information sample (training sample), using reinforcement learning, or supervised, or unsupervised methods for separate training; or it can be based on the baby coaxing to sleep information sample, using reinforcement learning, and jointly trained with the decision model for training. Specifically, the training material of the information fusion model (baby coaxing to sleep information sample) comes from the coaxing process data of multiple babies, and the baby coaxing process data is specifically at time = 0, T, 2T, 3T...nT, baby facial state information, limb information, external temperature, humidity, current time, etc., as well as the operating parameter information of the parent rocking the electric crib, and the baby's sleep evaluation at the current time t relative to the previous time tT. For example, the baby's sleep evaluation can be +1 if the baby is close to falling asleep, otherwise -1. The operating information (motor parameters) of the electric crib rocked by the parent can be captured by sensors, recorded, and stored as training samples for the reinforcement learning algorithm. Other baby-related information can also be recorded for future use. Multimodal fusion, based on the source characteristics of the information, can be used to combine feature fusion and decision fusion to process the multi-dimensional information and obtain the target baby's status information.

[0074] Step S300: Analyzing and processing the target baby status information based on a preset decision model to obtain a target sleep-coaxing decision plan, and controlling the crib to perform a sleep-coaxing operation for the baby in the crib according to the target sleep-coaxing decision plan;

[0075] Among them, the decision model is obtained according to the following steps: inputting the baby status information sample into the preset first model to be trained to obtain a predicted coaxing to sleep decision plan; secondly, after controlling the baby crib to execute the predicted coaxing to sleep decision plan, obtaining the updated baby status information sample; then determining the state change score value based on the baby status information sample and the updated baby status information sample; finally, judging whether the state change score value meets the preset first loss convergence condition. If the state change score value does not meet the preset first loss convergence condition, returning to the step of inputting the baby status information sample into the preset first model to be trained to obtain a predicted coaxing to sleep decision plan, until the state change score value meets the preset first loss convergence condition, obtaining a decision model that meets the accuracy condition; the baby status information sample is obtained by performing multimodal fusion processing on the baby coaxing to sleep information sample based on a preset information fusion model, and the first model to be trained is a reinforcement learning model.

[0076] As an example, the device analyzes and processes the target baby status information based on a preset decision model to obtain a target coaxing-to-sleep decision plan, and controls the crib to perform the coaxing-to-sleep operation for the baby in the crib according to the target coaxing-to-sleep decision plan, wherein the target coaxing-to-sleep decision plan refers to a related decision plan for controlling the crib to achieve coaxing-to-sleep. The crib realizes the coaxing-to-sleep function for the baby based on the corresponding motor operating parameters. The specific target coaxing-to-sleep decision plan can be the motor operating parameters. The device generates the corresponding motor operating parameters based on the preset decision model and the current baby status information, and controls the crib to execute the target coaxing-to-sleep decision plan to achieve automatic control of the baby's coaxing-to-sleep.

[0077] As an example, the decision model is obtained by iteratively training a preset first model to be trained using reinforcement learning, wherein reinforcement learning is a learning mechanism that learns how to map from state to behavior so as to maximize the reward obtained. The intelligent agent (decision model) needs to continuously experiment in the environment, and continuously optimize the state-behavior correspondence through the feedback (reward) given by the environment, wherein repeated experiments (iterative training) and delayed rewards (state change score value) are the two most important features of reinforcement learning. Therefore, the present application adopts a model training method of reinforcement learning, and trains a decision model by autonomously exploring the environment without the need for historical coaxing data to sleep. There are no other conditions or restrictions, and it is applicable to coaxing all babies to sleep, thereby improving the flexibility of crib control.

[0078] The present application provides a method for coaxing a baby to sleep. Compared with the related art, it is necessary to collect the baby's historical coaxing data in advance before the baby's coaxing data can be controlled. For the baby without collected coaxing data, the corresponding control cannot be achieved, and if there are few data records of the historical coaxing data, the control of the baby crib will be fixed and repetitive, which limits the control of the baby crib by this scheme, resulting in poor flexibility. In this application, multi-dimensional information related to the baby's sleep in the crib is obtained; based on a preset information fusion model, the multi-dimensional information is multimodally fused to obtain target baby status information; based on a preset decision model, the target baby status information is analyzed and processed to obtain a target coaxing decision plan, and the crib is controlled to perform the coaxing operation for the baby in the crib according to the target coaxing decision plan; wherein, the decision model is obtained according to the following steps: the baby status information sample is converted into a coaxing data sample; the baby crib is controlled to perform the coaxing operation for the baby in the crib according to the target coaxing decision plan; wherein, the decision model is obtained according to the following steps: This is input into the preset first model to be trained to obtain a predicted sleep-coaxing decision plan; secondly, after controlling the crib to execute the predicted sleep-coaxing decision plan, an updated baby state information sample is obtained; then based on the baby state information sample and the updated baby state information sample, a state change score value is determined; finally, it is judged whether the state change score value meets the preset first loss convergence condition. If the state change score value does not meet the preset first loss convergence condition, the step of returning to inputting the baby state information sample into the preset first model to be trained to obtain a predicted sleep-coaxing decision plan is performed until the state change score value meets the preset first loss convergence condition, and a decision model that meets the accuracy condition is obtained; the baby state information sample is obtained by performing multimodal fusion processing on the baby sleep-coaxing information sample based on a preset information fusion model, and the first model to be trained is a reinforcement learning model. That is, in this application, a reinforcement learning model training method is adopted, and the decision model algorithm is optimized based on the infant state change score before and after the execution of the predicted coaxing to sleep decision plan as instant feedback, so that the decision model can be trained by autonomously exploring the environment without the need for historical coaxing to sleep data. There are no other conditions or restrictions, and it is suitable for coaxing all infants to sleep, thereby improving the flexibility of crib control.

[0079] Based on the above first embodiment, the present application also provides another embodiment, referring to Figure 4 The method for coaxing a baby to sleep comprises the following steps A100-A500:

[0080] Step A100: When the user presses the relevant start button of the electric baby bed, the device starts the intelligent electric baby bed coaxing control program, and the initial state starts from the default setting;

[0081] Step A200: The device obtains the baby's current facial and limb information, as well as external information such as temperature, humidity, and light;

[0082] Step A300: Inputting the multiple pieces of information from step A200 into a pre-designed multimodal fusion model to obtain a comprehensive value for measuring the infant's sleeping state;

[0083] Step A400: Input the baby's sleeping state value from step A300 into the reinforcement learning model, calculate the baby crib control parameters, update the rocking frequency / amplitude / angle of the electric rocking chair, and continue the operation for T time;

[0084] Step A500: If the baby is asleep for a predetermined time and wakes up, the parent reminder operation is initiated. If the baby wakes up before the predetermined time or cries, the process returns to step A200 and subsequent steps.

[0085] As an example, the reinforcement learning model training method is used to train the decision model through autonomous exploration of the environment without the need for historical sleep-coaxing data. There are no other conditions or restrictions, and it is suitable for coaxing all babies to sleep, thereby improving the flexibility of crib control.

[0086] Based on the above first embodiment, the present application also provides another embodiment, referring to Figure 5 , the method for coaxing a baby to sleep comprises:

[0087] Before the step S100 of acquiring multi-dimensional information related to the baby sleeping in the crib, the method includes the following steps B100-B500:

[0088] Step B100, obtaining a sample of baby coaxing sleep information;

[0089] As an example, the baby coaxing to sleep information sample is related to the baby coaxing to sleep, and the sample used for model training is derived from the data of the coaxing process of multiple babies, wherein, with reference to Figure 5 The data of the baby coaxing to sleep process specifically includes the baby's facial state information, limb information, external temperature, humidity, current time, etc. at each time point time = 0, T, 2T, 3T...nT, as well as the operating parameter information of the parent shaking the electric crib, and the baby's sleep evaluation at the current time t relative to the previous time tT. That is, the baby coaxing to sleep information sample includes but is not limited to the baby's facial state information, limb information, external temperature information, external humidity information, current time information, etc.

[0090] As an example, the device may obtain baby sleep information samples by collecting corresponding information based on relevant sensors (such as cameras, sound sensors, temperature sensors, humidity sensors, clocks, etc.); or it may obtain the information by receiving baby sleep information samples uploaded by users.

[0091] It should be noted that the device can only obtain baby sleep information samples with the authorization of the user (or parent), and the user has been informed that the baby sleep information samples collected by the device are used to train the decision model, and the collected baby sleep information samples will not be used in other scenarios unrelated to this application.

[0092] Step B200, performing multimodal fusion processing on the infant sleep coaxing information sample based on a preset information fusion model to obtain an infant status information sample;

[0093] As an example, the device performs multimodal fusion processing on the baby sleeping information sample based on a preset information fusion model to obtain a baby status information sample. The processing method of this step refers to step S200 and is not repeated here.

[0094] Step B300: inputting the infant status information sample into a preset first model to be trained to obtain a prediction and decision scheme for coaxing the infant to sleep, wherein the first model to be trained is a reinforcement learning model;

[0095] As an example, the device inputs the infant status information sample into a preset first model to be trained to obtain a predicted sleep-coaxing decision plan, wherein the first model to be trained is a reinforcement learning model, which is the initial training model of the decision model. The first model to be trained has the ability to process the infant status information sample and obtain the corresponding predicted sleep-coaxing decision plan. Compared with the decision model, the only difference is the prediction accuracy. Specifically, the reinforcement learning model (the first model to be trained) operates on the infant status information sample to obtain new motor operation parameters (predicted sleep-coaxing decision plan).

[0096] Step B400, after controlling the crib to execute the predictive coaxing-to-sleep decision plan, obtaining an updated infant status information sample, and determining a status change score based on the infant status information sample and the updated infant status information sample;

[0097] As an example, after the device obtains a predicted sleep-coaxing decision plan based on the first model to be trained, it controls the crib to execute the predicted sleep-coaxing decision plan, wherein the parents need to wait for a fixed time T to coax the baby to sleep, but in the algorithm training, this T time is regarded as a sequential step and does not require actual waiting. After controlling the crib to execute the predicted sleep-coaxing decision plan, the environment related to the baby's sleep (the environment includes the baby's body information, external information, the baby's sleep state, etc.) changes accordingly, and the device re-acquires the baby's state information sample after the current change (that is, the updated baby state information sample), and determines the state change score value based on the baby state information sample and the updated baby state information sample, wherein the state change score value is the feedback (reward) given by the environment. Specifically, this application uses the evaluation of the baby's sleep state (good or bad) twice before and after (or multiple times in multiple time periods) as immediate feedback or reward.

[0098] Step B500, determine whether the state change score value meets the preset first loss convergence condition. If the state change score value does not meet the preset first loss convergence condition, return to the step of inputting the baby state information sample into the preset first model to be trained to obtain a step of predicting a sleep-coaxing decision plan, until the state change score value meets the preset first loss convergence condition, and a decision model that meets the accuracy condition is obtained.

[0099] As an example, the device determines whether the state change score value meets the preset first loss convergence condition. If the state change score value does not meet the preset first loss convergence condition, it returns to the step of inputting the infant state information sample into the preset first model to be trained to obtain a step of predicting a sleep-coaxing decision plan, until the state change score value meets the preset first loss convergence condition, and a decision model that meets the accuracy condition is obtained, that is, the present application uses the state change score value as an immediate feedback or reward in reinforcement learning to optimize the algorithm. Specifically, if the state change score value does not meet the preset first loss convergence condition, it means that the relevant parameters in the algorithm of the current first model to be trained have not reached the convergence condition, then the device optimizes the relevant parameters in the algorithm of the current first model to be trained according to the state change score value, wherein the parameter optimization method can be to update the parameters in a gradient ascent manner, and to transfer the gradient from the state value back to the multimodal fusion model to update the parameters of each model therein. Based on the first model to be trained of the optimized algorithm, repeat the above steps B100-B500 until the state change score value meets the preset first loss convergence condition, which means that the relevant parameters in the algorithm of the current first model to be trained have reached the convergence condition, and the reinforcement learning model training is completed to obtain a decision model that meets the accuracy conditions.

[0100] As an example, this application introduces a reinforcement learning algorithm and trains a decision model. The device makes decisions and controls the electric crib based on the decision model, which is consistent with the process of human cognitive learning and the parents' observation of the baby's body, face, and external information to determine the rate and direction of rocking the crib, thereby helping the baby fall asleep. In order to enable the training and use of reinforcement learning algorithms, the key lies in how to involve incentives or feedback. This application proposes to use two sleep status (good or bad) evaluations before and after as immediate feedback or rewards to optimize the algorithm, so as to achieve the training of the decision model through autonomous exploration of the environment without the need for historical sleep data. There are no other conditions and restrictions, and it is applicable to all babies' sleep, thereby improving the flexibility of crib control.

[0101] Before the step B200 of performing multimodal fusion processing on the infant sleep coaxing information sample based on a preset information fusion model to obtain an infant status information sample, the method includes the following steps C100:

[0102] Step C100, performing random information erasure processing on the baby coaxing to sleep information under the baby status information sample to obtain an erased baby coaxing to sleep information sample;

[0103] The erased baby coaxing to sleep information samples include some baby coaxing to sleep information samples of the baby status information samples, or the erased baby coaxing to sleep information samples include all baby coaxing to sleep information samples of the baby status information samples.

[0104] As an example, in the process of training the decision model, the device performs random information erasure processing on the baby coaxing information sample under the baby state information sample before performing multimodal fusion processing on the baby coaxing information sample based on the preset information fusion model to obtain the erased baby coaxing information sample, wherein the erased baby coaxing information sample includes part of the baby coaxing information sample of the baby state information sample, or the erased baby coaxing information sample includes all the baby coaxing information samples of the baby state information sample, that is, the random information erasure processing can be to erase part of the information or not to erase any information, but there will be no situation where all the information is erased. Because some information is inevitably missing when collecting samples of the baby coaxing process (such as time information is not recorded), this application designs a module for random information erasure to enable the algorithm model to perform intelligent control even when some information is missing, thereby enhancing the robustness of the algorithm, so that when the algorithm model encounters some information that cannot be obtained in product applications, it can also stably and reliably perform the corresponding information processing work.

[0105] Based on the above-mentioned first and second embodiments, the present application further provides another embodiment, wherein the method for coaxing a baby to sleep comprises:

[0106] After the step B400 of obtaining an updated infant status information sample after controlling the infant crib to execute the predictive coaxing-to-sleep decision plan, and determining a status change score based on the infant status information sample and the updated infant status information sample, the method includes the following steps D100-D300:

[0107] Step D100, after controlling the crib to execute the predicted coaxing-to-sleep decision plan, obtaining a coaxing-to-sleep result for the baby;

[0108] As an example, the baby coaxing sleep result refers to the result of whether the baby has successfully fallen asleep. After controlling the crib to execute the predicted coaxing sleep decision plan, the environment related to the baby's falling asleep (the environment includes the baby's own information, external information, the baby's sleeping status, etc.) changes accordingly, and the device re-acquires the baby coaxing sleep information sample and the updated baby status information sample after the current change. The device determines the baby coaxing sleep result based on the baby coaxing sleep information sample and / or the updated baby status information sample. The baby coaxing sleep result can be analyzed according to the judgment method (or rules) related to whether the baby has successfully fallen asleep, and it is determined whether the current baby has successfully fallen asleep to obtain the baby coaxing sleep result.

[0109] Step D200, determining whether the state change score value satisfies a preset first loss convergence condition, and determining whether the baby sleeping result satisfies a preset second loss convergence condition;

[0110] As an example, the device determines whether the state change score value meets the preset first loss convergence condition, and determines whether the baby coaxing to sleep result meets the preset second loss convergence condition. This application proposes to use the two previous sleep state (good or bad) evaluations and the baby coaxing to sleep results as instant feedback or rewards. In each round of iteration cycle, the device obtains the state change score value and the baby coaxing to sleep result, and determines whether the state change score value meets the preset first loss convergence condition, and determines whether the baby coaxing to sleep result meets the preset second loss convergence condition. In this way, it is combined to determine whether the model converges.

[0111] Step D300: If the state change score value does not meet the preset first loss convergence condition or the baby coaxing to sleep result does not meet the preset second loss convergence condition, return to the step of inputting the baby state information sample into the preset first model to be trained to obtain a predicted coaxing to sleep decision plan, until the state change score value meets the preset first loss convergence condition and the baby coaxing to sleep result meets the preset second loss convergence condition, and a decision model that meets the accuracy condition is obtained.

[0112] As an example, if the state change score value does not meet the preset first loss convergence condition or the baby coaxing to sleep result does not meet the preset second loss convergence condition, the device returns to the step of inputting the baby state information sample into the preset first model to be trained to obtain a predicted coaxing to sleep decision plan, until the state change score value meets the preset first loss convergence condition and the baby coaxing to sleep result meets the preset second loss convergence condition, and a decision model that meets the accuracy condition is obtained. Specifically, if the state change score value does not meet the preset first loss convergence condition or the baby coaxing to sleep result does not meet the preset second loss convergence condition, it means that the relevant parameters in the algorithm of the current first model to be trained have not reached the convergence condition. The device optimizes the relevant parameters in the algorithm of the current first model to be trained according to the state change score value and the baby coaxing to sleep result, and repeats the above model training steps based on the first model to be trained of the optimized algorithm until the state change score value meets the preset first loss convergence condition and the baby coaxing to sleep result meets the preset second loss convergence condition. It means that the relevant parameters in the algorithm of the current first model to be trained have reached the convergence condition, the reinforcement learning model training is completed, and a decision model that meets the accuracy conditions is obtained.

[0113] As an example, this application proposes to use the sleep status (good or bad) evaluations and the results of coaxing the baby to sleep twice as immediate feedback or rewards, that is, combining the sleep status evaluation and the results of coaxing the baby to sleep to judge the loss convergence of the model. Only when the state change score value and the baby coaxing result of the current model iteration cycle meet the corresponding loss convergence conditions at the same time, the reinforcement learning model training is completed, and a decision model that meets the accuracy conditions is obtained, thereby improving the prediction accuracy of the decision model.

[0114] Based on the above-mentioned first, second and third embodiments, the present application further provides another embodiment, wherein the method for coaxing a baby to sleep comprises:

[0115] Before the step S100 of acquiring multi-dimensional information related to the baby sleeping in the crib, the method further includes the following steps E100-E200:

[0116] Step E100, obtaining a sample of baby sleep-coaxing information;

[0117] As an example, this step refers to the above step B100 and will not be repeated here.

[0118] Step E200: Based on the infant sleep-coaxing information sample, a preset second to-be-trained model and a preset first to-be-trained model are jointly trained using a reinforcement learning method to obtain an information fusion model and a decision model that meet accuracy requirements;

[0119] The second model to be trained is the initial training model of the information fusion model, and the first model to be trained is the initial training model of the decision model.

[0120] As an example, since both the information fusion model and the decision model use the two previous sleep status (good or bad) evaluations as immediate feedback or rewards, the information fusion model and the decision model can be jointly trained. Specifically, the device uses reinforcement learning based on the baby sleep information sample to jointly train the preset second model to be trained and the preset first model to be trained to obtain an information fusion model and a decision model that meet the accuracy conditions, wherein the second model to be trained is the initial training model of the information fusion model, and the first model to be trained is the initial training model of the decision model.

[0121] Specifically, the step E200 includes the following steps E210-E240:

[0122] Step E210: inputting the baby sleeping information sample into a preset second to-be-trained model to obtain a baby status information sample;

[0123] As an example, the device inputs the infant sleep information sample into a preset second to-be-trained model to obtain an infant status information sample. The second to-be-trained model is the initial training model for the information fusion model. The second to-be-trained model is capable of processing the infant sleep information sample and obtaining the corresponding infant status information sample. Compared to the information fusion model, the second to-be-trained model only differs in prediction accuracy. Specifically, the reinforcement learning model (the second to-be-trained model) performs multimodal fusion processing on the infant sleep information sample to obtain the infant status information sample.

[0124] Step E220: Inputting the infant status information sample into a preset first model to be trained to obtain a prediction and decision scheme for coaxing the infant to sleep, wherein both the first model to be trained and the second model to be trained are reinforcement learning models;

[0125] As an example, this step refers to the above step B300 and will not be repeated here.

[0126] Step E230: After controlling the crib to execute the predictive coaxing-to-sleep decision plan, obtaining an updated infant status information sample, and determining a status change score based on the infant status information sample and the updated infant status information sample;

[0127] As an example, this step refers to the above step B400 and will not be repeated here.

[0128] Step E240, determine whether the state change score value meets the preset first loss convergence condition. If the state change score value does not meet the preset loss convergence condition, return to the step of inputting the baby sleep information sample into the preset second model to be trained to obtain the baby state information sample, until the state change score value meets the preset first loss convergence condition, and obtain an information fusion model and decision model that meet the accuracy conditions.

[0129] As an example, the device determines whether the state change score value meets the preset first loss convergence condition. If the state change score value does not meet the preset loss convergence condition, it returns to the step of inputting the baby sleep information sample into the preset second model to be trained to obtain the baby state information sample until the state change score value meets the preset first loss convergence condition, thereby obtaining an information fusion model and a decision model that meet the accuracy conditions. Both the information fusion model and the decision model use the two previous sleep state (good or bad) evaluations as immediate feedback or rewards to optimize the second model to be trained and the first model to be trained. Specifically, if the state change score value does not meet the preset first loss convergence condition, it means that the relevant parameters in the algorithms of the current first model to be trained and the second model to be trained have not reached the convergence condition. Then, the device optimizes the relevant parameters in the algorithms of the current first model to be trained and the second model to be trained according to the state change score value, and repeats the above joint model training based on the first model to be trained and the second model to be trained of the optimized algorithm until the state change score value meets the preset first loss convergence condition, which means that the relevant parameters in the algorithms of the current first model to be trained and the second model to be trained have reached the convergence condition, completing the reinforcement learning model training and obtaining an information fusion model and a decision model that meet the accuracy conditions.

[0130] This application also provides a baby sleeping device, referring to Figure 3 , the baby sleeping device comprises:

[0131] an acquisition module 10 for acquiring multi-dimensional information related to the baby's sleep in the crib;

[0132] A fusion module 20 is configured to perform multimodal fusion processing on the multi-dimensional information based on a preset information fusion model to obtain target infant status information;

[0133] an execution module 30 for analyzing and processing the target infant status information based on a preset decision model to obtain a target coaxing-to-sleep decision plan, and controlling the crib to perform a coaxing-to-sleep operation for the infant in the crib according to the target coaxing-to-sleep decision plan;

[0134] Among them, the decision model is obtained according to the following steps: inputting the baby status information sample into the preset first model to be trained to obtain a predicted coaxing to sleep decision plan; secondly, after controlling the baby crib to execute the predicted coaxing to sleep decision plan, obtaining the updated baby status information sample; then determining the state change score value based on the baby status information sample and the updated baby status information sample; finally, judging whether the state change score value meets the preset first loss convergence condition. If the state change score value does not meet the preset first loss convergence condition, returning to the step of inputting the baby status information sample into the preset first model to be trained to obtain a predicted coaxing to sleep decision plan, until the state change score value meets the preset first loss convergence condition, obtaining a decision model that meets the accuracy condition; the baby status information sample is obtained by performing multimodal fusion processing on the baby coaxing to sleep information sample based on a preset information fusion model, and the first model to be trained is a reinforcement learning model.

[0135] And / or, the baby coaxing sleep device further includes: an information erasing module, configured to perform random information erasing processing on the baby coaxing sleep information under the baby status information sample to obtain an erased baby coaxing sleep information sample; wherein the erased baby coaxing sleep information sample includes a portion of the baby status information sample, or the erased baby coaxing sleep information sample includes all of the baby coaxing sleep information samples of the baby status information sample;

[0136] And / or, the baby coaxing sleep device further includes: a coaxing sleep result acquisition module, used to obtain the baby coaxing sleep result after controlling the baby crib to execute the predicted coaxing sleep decision plan; a judgment module, used to judge whether the state change score value meets the preset first loss convergence condition, and judge whether the baby coaxing sleep result meets the preset second loss convergence condition; a second iterative training module, used to return to the step of inputting the baby state information sample into the preset first to-be-trained model to obtain the predicted coaxing sleep decision plan if the state change score value does not meet the preset first loss convergence condition or the baby coaxing sleep result does not meet the preset second loss convergence condition, until the state change score value meets the preset first loss convergence condition and the baby coaxing sleep result meets the preset second loss convergence condition, thereby obtaining a decision model that meets the accuracy condition;

[0137] And / or, the infant coaxing device to sleep further includes: a joint training module for jointly training a preset second model to be trained and a preset first model to be trained using a reinforcement learning method based on the infant coaxing information sample to obtain an information fusion model and a decision model that meet accuracy requirements; wherein the second model to be trained is an initial training model for the information fusion model, and the first model to be trained is an initial training model for the decision model;

[0138] And / or, the joint training module includes: a second state information sample determination module, used to input the baby sleep coaxing information sample into a preset second model to be trained to obtain a baby state information sample; a second scheme prediction module, used to input the baby state information sample into a preset first model to be trained to obtain a predicted sleep coaxing decision scheme, wherein the first model to be trained and the second model to be trained are both reinforcement learning models; a second change scoring module, used to obtain an updated baby state information sample after controlling the baby crib to execute the predicted sleep coaxing decision scheme, and determine a state change score value based on the baby state information sample and the updated baby state information sample; a third iterative training module, used to determine whether the state change score value meets a preset first loss convergence condition. If the state change score value does not meet the preset loss convergence condition, return to the step of inputting the baby sleep coaxing information sample into the preset second model to be trained to obtain the baby state information sample, until the state change score value meets the preset first loss convergence condition, and obtain an information fusion model and decision model that meet the accuracy condition.

[0139] The specific implementation of the baby sleeping device of the present application is basically the same as the embodiments of the above-mentioned baby sleeping method, and will not be repeated here.

[0140] Reference Figure 1 , Figure 1 It is a schematic diagram of the terminal structure of the hardware operating environment involved in the embodiment of the present application.

[0141] like Figure 1 As shown, the terminal may include: a processor 1001, such as a CPU, a network interface 1004, a user interface 1003, a memory 1005, and a communication bus 1002. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display), an input unit such as a keyboard (Keyboard), and the user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory 1005 may be a high-speed RAM memory or a stable memory (non-volatile memory), such as a disk memory. The memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0142] Optionally, the baby coaxing device may further include a rectangular user interface, a network interface, a camera, an RF (Radio Frequency) circuit, a sensor, an audio circuit, a WiFi module, and the like. The rectangular user interface may include a display and an input submodule such as a keyboard. Optionally, the rectangular user interface may also include a standard wired interface and a wireless interface. The network interface may optionally include a standard wired interface and a wireless interface (such as a WiFi interface).

[0143] Those skilled in the art will understand that Figure 1 The structure of the baby lulling device shown in the figure does not constitute a limitation on the baby lulling device, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0144] like Figure 1 As shown, the memory 1005, which serves as a storage medium, may include an operating system, a network communication module, and a baby lulling program. The operating system is a program that manages and controls the hardware and software resources of the baby lulling device and supports the operation of the baby lulling program and other software and / or programs. The network communication module is used to enable communication between the various components within the memory 1005, as well as communication with other hardware and software in the baby lulling system.

[0145] exist Figure 1 In the baby coaxing device shown, the processor 1001 is used to execute the baby coaxing program stored in the memory 1005 to implement the steps of any of the above-mentioned baby coaxing methods.

[0146] The specific implementation of the baby coaxing device to sleep in the present application is basically the same as the embodiments of the above-mentioned baby coaxing method to sleep, and will not be repeated here.

[0147] The present application further provides a storage medium, wherein a program for implementing a method for lulling an infant to sleep is stored on the storage medium. The program for implementing a method for lulling an infant to sleep is executed by a processor to implement the following method for lulling an infant to sleep:

[0148] Obtain multidimensional information related to infant sleep in the crib;

[0149] Based on a preset information fusion model, performing multimodal fusion processing on the multi-dimensional information to obtain target infant status information;

[0150] Analyzing and processing the target infant status information based on a preset decision model to obtain a target sleep-coaxing decision plan, and controlling the crib to perform a sleep-coaxing operation for the infant in the crib according to the target sleep-coaxing decision plan;

[0151] Among them, the decision model is obtained according to the following steps: inputting the baby status information sample into the preset first model to be trained to obtain a predicted coaxing to sleep decision plan; secondly, after controlling the baby crib to execute the predicted coaxing to sleep decision plan, obtaining the updated baby status information sample; then determining the state change score value based on the baby status information sample and the updated baby status information sample; finally, judging whether the state change score value meets the preset first loss convergence condition. If the state change score value does not meet the preset first loss convergence condition, returning to the step of inputting the baby status information sample into the preset first model to be trained to obtain a predicted coaxing to sleep decision plan, until the state change score value meets the preset first loss convergence condition, obtaining a decision model that meets the accuracy condition; the baby status information sample is obtained by performing multimodal fusion processing on the baby coaxing to sleep information sample based on a preset information fusion model, and the first model to be trained is a reinforcement learning model.

[0152] Optionally, the multi-dimensional information includes infant body information, time information, and environmental information related to infant sleep.

[0153] Optionally, before the step of inputting the infant status information sample into a preset first to-be-trained model to obtain a predicted sleep-coaxing decision plan, the method includes:

[0154] performing random information erasure processing on the baby coaxing to sleep information under the baby status information sample to obtain an erased baby coaxing to sleep information sample;

[0155] The erased baby coaxing to sleep information samples include some baby coaxing to sleep information samples of the baby status information samples, or the erased baby coaxing to sleep information samples include all baby coaxing to sleep information samples of the baby status information samples.

[0156] Optionally, after the step of determining a status change score value based on the infant status information sample and the updated infant status information sample, the method includes:

[0157] After controlling the crib to execute the predicted coaxing-to-sleep decision plan, obtaining a coaxing-to-sleep result for the baby;

[0158] The step of determining whether the state change score value satisfies a preset first loss convergence condition, and if the state change score value does not satisfy the preset first loss convergence condition, returning to the step of inputting the infant state information sample into a preset first to-be-trained model to obtain a prediction sleep decision plan, until the state change score value satisfies the preset first loss convergence condition and a decision model that satisfies the accuracy condition is obtained, includes:

[0159] Determining whether the state change score value satisfies a preset first loss convergence condition, and determining whether the baby sleeping result satisfies a preset second loss convergence condition;

[0160] If the state change score value does not meet the preset first loss convergence condition or the baby sleeping result does not meet the preset second loss convergence condition, return to the step of inputting the baby state information sample into the preset first to-be-trained model to obtain the step of predicting the sleeping decision plan, until the state change score value meets the preset first loss convergence condition and the baby sleeping result meets the preset second loss convergence condition, and a decision model that meets the accuracy conditions is obtained.

[0161] Optionally, before the step of obtaining multi-dimensional information related to infant sleep, the method includes:

[0162] Get sample baby sleep information;

[0163] Based on the infant sleep coaxing information sample, the preset second to-be-trained model and the preset first to-be-trained model are jointly trained using a reinforcement learning method to obtain an information fusion model and a decision model that meet the accuracy conditions;

[0164] The second model to be trained is the initial training model of the information fusion model, and the first model to be trained is the initial training model of the decision model.

[0165] Optionally, the step of jointly training the preset second model to be trained and the preset first model to be trained using reinforcement learning based on the infant coaxing-to-sleep information sample to obtain an information fusion model and a decision model that meet accuracy requirements includes:

[0166] Inputting the infant sleep coaxing information sample into a preset second to-be-trained model to obtain an infant status information sample;

[0167] Inputting the infant status information sample into a preset first model to be trained to obtain a prediction and decision scheme for coaxing the infant to sleep, wherein the first model to be trained and the second model to be trained are both reinforcement learning models;

[0168] After controlling the crib to execute the predictive coaxing-to-sleep decision plan, obtaining an updated infant status information sample, and determining a status change score value based on the infant status information sample and the updated infant status information sample;

[0169] Determine whether the state change score value meets the preset first loss convergence condition. If the state change score value does not meet the preset loss convergence condition, return to the step of inputting the baby sleep information sample into the preset second model to be trained to obtain the baby state information sample, until the state change score value meets the preset first loss convergence condition, and obtain an information fusion model and a decision model that meet the accuracy conditions.

[0170] The specific implementation of the storage medium of the present application is basically the same as the embodiments of the above-mentioned method for coaxing a baby to sleep, and will not be repeated here.

[0171] The present application also provides a computer program product, including a computer program, which implements the steps of the above-mentioned method for coaxing an infant to sleep when executed by a processor.

[0172] The specific implementation of the computer program product of the present application is basically the same as the embodiments of the above-mentioned method for coaxing a baby to sleep, and will not be repeated here.

[0173] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or apparatus comprising the element.

[0174] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0175] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for enabling a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0176] The above are only preferred embodiments of the present application and do not limit the scope of protection of the present application. Any equivalent structure or equivalent process transformation made using the contents of the description and drawings of this application, or directly or indirectly applied in other related technical fields, are also included in the scope of protection of the present application.

Claims

1. A method for coaxing a baby to sleep, characterized in that: include: Obtain multidimensional information related to infant sleep in the crib; Based on a preset information fusion model, performing multimodal fusion processing on the multi-dimensional information to obtain target infant status information; Analyzing and processing the target infant status information based on a preset decision model to obtain a target sleep-coaxing decision plan, and controlling the crib to perform a sleep-coaxing operation for the infant in the crib according to the target sleep-coaxing decision plan; The decision model is obtained according to the following steps: inputting a sample of infant status information into a preset first to-be-trained model to obtain a predicted sleep-coaxing decision plan; Secondly, after controlling the baby crib to execute the predicted coaxing to sleep decision plan, an updated baby state information sample is obtained; then, based on the baby state information sample and the updated baby state information sample, a state change score value is determined; finally, it is judged whether the state change score value meets the preset first loss convergence condition. If the state change score value does not meet the preset first loss convergence condition, the step of inputting the baby state information sample into the preset first model to be trained to obtain the predicted coaxing to sleep decision plan is returned until the state change score value meets the preset first loss convergence condition, and a decision model that meets the accuracy condition is obtained; the baby state information sample is obtained by performing multimodal fusion processing on the baby coaxing to sleep information sample based on a preset information fusion model, and the first model to be trained is a reinforcement learning model.

2. The method for coaxing a baby to sleep according to claim 1, wherein: The multi-dimensional information includes infant body information, time information, and environmental information related to the infant's sleep.

3. The method for coaxing a baby to sleep according to claim 1, wherein: Before the step of inputting the infant status information sample into the preset first to-be-trained model to obtain a predicted sleep-coaxing decision plan, the method includes: performing random information erasure processing on the baby coaxing to sleep information under the baby status information sample to obtain an erased baby coaxing to sleep information sample; The erased baby coaxing to sleep information samples include some baby coaxing to sleep information samples of the baby status information samples, or the erased baby coaxing to sleep information samples include all baby coaxing to sleep information samples of the baby status information samples.

4. The method for coaxing a baby to sleep according to claim 1, wherein: After the step of determining a status change score based on the infant status information sample and the updated infant status information sample, the method includes: After controlling the crib to execute the predicted coaxing-to-sleep decision plan, obtaining a coaxing-to-sleep result for the baby; The step of determining whether the state change score value satisfies a preset first loss convergence condition, and if the state change score value does not satisfy the preset first loss convergence condition, returning to the step of inputting the infant state information sample into a preset first to-be-trained model to obtain a prediction sleep decision plan, until the state change score value satisfies the preset first loss convergence condition and a decision model that satisfies the accuracy condition is obtained, includes: Determining whether the state change score value satisfies a preset first loss convergence condition, and determining whether the baby sleeping result satisfies a preset second loss convergence condition; If the state change score value does not meet the preset first loss convergence condition or the baby sleeping result does not meet the preset second loss convergence condition, return to the step of inputting the baby state information sample into the preset first to-be-trained model to obtain the step of predicting the sleeping decision plan, until the state change score value meets the preset first loss convergence condition and the baby sleeping result meets the preset second loss convergence condition, and a decision model that meets the accuracy conditions is obtained.

5. The method for coaxing a baby to sleep according to claim 1, wherein: Before the step of obtaining multi-dimensional information related to infant sleep, the method includes: Get sample baby sleep information; Based on the infant sleep coaxing information sample, the preset second to-be-trained model and the preset first to-be-trained model are jointly trained using a reinforcement learning method to obtain an information fusion model and a decision model that meet the accuracy conditions; The second model to be trained is the initial training model of the information fusion model, and the first model to be trained is the initial training model of the decision model.

6. The method for coaxing a baby to sleep according to claim 5, wherein: The step of jointly training the preset second to-be-trained model and the preset first to-be-trained model using reinforcement learning based on the infant sleep-coaxing information sample to obtain an information fusion model and a decision model that meet the accuracy conditions includes: Inputting the infant sleep coaxing information sample into a preset second to-be-trained model to obtain an infant status information sample; Inputting the infant status information sample into a preset first model to be trained to obtain a prediction and decision scheme for coaxing the infant to sleep, wherein the first model to be trained and the second model to be trained are both reinforcement learning models; After controlling the crib to execute the predictive coaxing-to-sleep decision plan, obtaining an updated infant status information sample, and determining a status change score value based on the infant status information sample and the updated infant status information sample; Determine whether the state change score value meets the preset first loss convergence condition. If the state change score value does not meet the preset loss convergence condition, return to the step of inputting the baby sleep information sample into the preset second model to be trained to obtain the baby state information sample, until the state change score value meets the preset first loss convergence condition, and obtain an information fusion model and a decision model that meet the accuracy conditions.

7. A baby sleeping device, characterized in that: The baby sleeping device comprises: an acquisition module, for acquiring multi-dimensional information related to the baby's sleep in the crib; a fusion module, configured to perform multimodal fusion processing on the multi-dimensional information based on a preset information fusion model to obtain target infant status information; an execution module, configured to analyze and process the target infant status information based on a preset decision model to obtain a target sleep-coaxing decision plan, and control the crib to perform a sleep-coaxing operation on the infant in the crib according to the target sleep-coaxing decision plan; Among them, the decision model is obtained according to the following steps: inputting the baby status information sample into the preset first model to be trained to obtain a predicted coaxing to sleep decision plan; secondly, after controlling the baby crib to execute the predicted coaxing to sleep decision plan, obtaining the updated baby status information sample; then determining the state change score value based on the baby status information sample and the updated baby status information sample; finally, judging whether the state change score value meets the preset first loss convergence condition. If the state change score value does not meet the preset first loss convergence condition, returning to the step of inputting the baby status information sample into the preset first model to be trained to obtain a predicted coaxing to sleep decision plan, until the state change score value meets the preset first loss convergence condition, obtaining a decision model that meets the accuracy condition; the baby status information sample is obtained by performing multimodal fusion processing on the baby coaxing to sleep information sample based on a preset information fusion model, and the first model to be trained is a reinforcement learning model.

8. The baby sleeping device according to claim 7, characterized in that: The baby sleeping device also includes: an information erasing module, configured to perform random information erasure processing on the baby coaxing information under the baby status information sample to obtain an erased baby coaxing information sample; wherein the erased baby coaxing information sample includes a portion of the baby coaxing information sample of the baby status information sample, or the erased baby coaxing information sample includes all of the baby coaxing information samples of the baby status information sample; And / or, the baby coaxing sleep device further includes: a coaxing sleep result acquisition module, used to obtain the baby coaxing sleep result after controlling the baby crib to execute the predicted coaxing sleep decision plan; a judgment module, used to judge whether the state change score value meets the preset first loss convergence condition, and judge whether the baby coaxing sleep result meets the preset second loss convergence condition; a second iterative training module, used to return to the step of inputting the baby state information sample into the preset first to-be-trained model to obtain the predicted coaxing sleep decision plan if the state change score value does not meet the preset first loss convergence condition or the baby coaxing sleep result does not meet the preset second loss convergence condition, until the state change score value meets the preset first loss convergence condition and the baby coaxing sleep result meets the preset second loss convergence condition, thereby obtaining a decision model that meets the accuracy condition; And / or, the infant coaxing device to sleep further includes: a joint training module for jointly training a preset second model to be trained and a preset first model to be trained using a reinforcement learning method based on the infant coaxing information sample to obtain an information fusion model and a decision model that meet accuracy requirements; wherein the second model to be trained is an initial training model for the information fusion model, and the first model to be trained is an initial training model for the decision model; And / or, the joint training module includes: a second state information sample determination module, used to input the baby sleep coaxing information sample into a preset second model to be trained to obtain a baby state information sample; a second scheme prediction module, used to input the baby state information sample into a preset first model to be trained to obtain a predicted sleep coaxing decision scheme, wherein the first model to be trained and the second model to be trained are both reinforcement learning models; a second change scoring module, used to obtain an updated baby state information sample after controlling the baby crib to execute the predicted sleep coaxing decision scheme, and determine a state change score value based on the baby state information sample and the updated baby state information sample; a third iterative training module, used to determine whether the state change score value meets a preset first loss convergence condition. If the state change score value does not meet the preset loss convergence condition, return to the step of inputting the baby sleep coaxing information sample into the preset second model to be trained to obtain the baby state information sample, until the state change score value meets the preset first loss convergence condition, and obtain an information fusion model and decision model that meet the accuracy condition.

9. A baby sleeping device, characterized in that: The baby-sleeping device comprises: a memory, a processor, and a program stored in the memory for implementing the baby-sleeping method. The memory is used to store a program for implementing a method for coaxing a baby to sleep; The processor is configured to execute a program for implementing the method for lulling an infant to sleep, so as to implement the steps of the method for lulling an infant to sleep as claimed in any one of claims 1 to 6.

10. A storage medium, characterized in that: The storage medium stores a program for implementing a method for coaxing a baby to sleep, and the program for implementing a method for coaxing a baby to sleep is executed by a processor to implement the steps of the method for coaxing a baby to sleep as claimed in any one of claims 1 to 6.

Images