JP2002304615A - Massage machine - Google Patents

Abstract

PROBLEM TO BE SOLVED: To place a massage machine in massaging operation with the optimum massage pattern and to perform massaging operation with the optimum pattern by varying the massage pattern in real time even when a user feels comfortable with a different body part. SOLUTION: A brain-wave signal detected by an Ag-AgCl electrode 4 is amplified by a signal processing part 5 and converted into a digital signal, which is inputted to a computer 6. The computer estimates massage effect according to the brain-wave signal and optimizes the massage pattern according to the estimation result. Namely, the brain-wave signal is processed by fast Fourier transformation to calculate a power spectrum. The obtained power spectrum is inputted to a neural network 62 to estimate comfortableness. From the estimated value of the comfortableness which is thus obtained, the optimum massage pattern is found by an enhanced learning function 63 by using an enhanced learning method which employ an action determination system for a lottery by random numbers.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a massage machine for treating or relaxing stiff shoulders.

[0002]

2. Description of the Related Art There are chair-type massage machines having a reclining mechanism and bed-type massage machines that are used while the user is lying on his / her back, for the purpose of treating stiff shoulders and the like and for providing a relaxation effect. These massage machines use a massage ball type in which one or two or more pairs of massage balls are mechanically operated by a crank mechanism due to the principle of massage stimulation, and push the human body by contraction of an air bag installed inside the massage machine. It is classified into a pneumatic type that squeezes or pinches, and a vibratory type that stimulates the human body with vibrations obtained by rotating a motor with an eccentric weight. High-performance products combining these have also been put to practical use.

[0003] Many of these massage machines have a remote control device for the user to determine the massage parameters such as the position, type and intensity of the massage stimulus, and the user of the massage machine operates the remote control. By doing so, you can receive your favorite massage.

[0004] Some of the massage machines, which are relatively expensive and have high functions, can finely move the position to be stimulated by remote control operation so as to hit a portion having a high massage effect in the human body, so-called acupoint. further,
Some have the function of automatically massaging various parts of the human body, such as the neck, shoulders, and waist, according to a pre-programmed procedure. This function is called a recommended course or an automatic course, and has a massage pattern that many users will find comfortable.

[0005] However, these massage machines require the user to adjust the position of the massage stimulus every time, and the operation is complicated and troublesome. Therefore, Japanese Patent Laid-Open No. 5-30
In Japanese Patent No. 0926, a band with an electrode for measuring electroencephalogram is attached to the head of a user of a massage machine, and massage is performed while detecting whether or not an alpha wave (an electroencephalogram of about 10 Hz) among the electroencephalogram of the user increases. Then, the massage pattern in which the α-wave becomes larger is set as the optimum pattern, and the massage operation is performed with the pattern.

[0006]

However, Japanese Patent Application Laid-Open No. 5-3
In the system disclosed in Japanese Patent No. 00926, the peak of the α wave among the brain waves is obtained by brain wave measurement, and the degree of relaxation of the user is estimated from the magnitude thereof. However, it is generally known that, for example, when the eyes are closed, the α-wave peak is higher than when the eyes are open, and it is considered that the degree of relaxation cannot be estimated only by the α-wave peak. Also, since the effect of the massage is detected almost in real time, even if it is possible to massage with a more effective massage pattern, there is no specific means for obtaining an optimal massage pattern, and there is no description. Therefore, there is no certainty that the obtained massage pattern is the optimal massage pattern, and there has been a problem that the massage operation cannot always be performed with the optimal massage pattern.

Therefore, according to the present invention, the massage operation can be performed with the optimal massage pattern, and even when the place where the user feels comfortable changes, the massage pattern can be changed in real time to perform the massage operation with the optimal massage pattern. Provide a massage machine.

[0008]

According to the first aspect of the present invention,
In a massage machine having a plurality of parameters for determining operating conditions, a massage for estimating the effect of a massage on a user when used based on physiological indexes of the user and / or based on evaluation information input from the user. Effect estimation means, and learning means for learning a massage pattern by increasing or decreasing the reward intensity used for reinforcement learning for each state of each parameter in response to an increase or decrease in the evaluation result of the massage effect estimated by the massage effect estimation means. And massage control means for performing massage control with the parameters learned by the learning means.

According to a second aspect of the present invention, in the massage machine according to the first aspect, the learning in the learning means uses the ratio of the reward intensity for each parameter state as a probability, and determines the state of each parameter by random selection. Is to decide.

[0010] According to a third aspect of the present invention, in the massage machine according to the first or second aspect, the value for increasing or decreasing the reward strength of the learning in the learning means is the value of the massage effect evaluation result estimated by the massage effect estimating means. The purpose is to integrate at least the reciprocal of the variation in the evaluation result with the increase / decrease amount.

[0011]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a view showing the overall configuration of a massage machine, wherein 1 is a massage chair, 2 is a user who sits on the massage chair 1 and receives a massage, 3 is a joystick operated by the user 2, and 4 is a user's 2 joystick. An Ag-AgCl electrode attached to the head to measure EEG
A signal processing unit 6 that takes in a signal from the g-AgCl electrode 4 and performs predetermined processing is a computer that drives and controls the massage chair 1 based on information from the signal processing unit 5.

The massage chair 1 has a plurality of airbags inside, and performs a massage operation by contracting the airbags at a constant cycle. The strength is adjusted according to the bulge of the airbags. It has a function called pulse operation that intermittently changes the inflation speed of the airbag. Then, the airbag shrinks to stimulate the human body, so that remission of discipline and pain and a feeling of relaxation can be selected. The airbags are installed at positions corresponding to the neck / shoulder, back, back muscles, waist, buttocks, thighs, legs (calf), and the like, and these places are sequentially (automatic course) or desired by the user. A specific position can be continuously stimulated.

Therefore, the massage chair 1 has a neck /
Shoulder, back, back muscles, waist, buttocks, thighs can massage seven parts of the legs, only legs have the function of simultaneous legs that can be operated even when other parts are operating I have.

There are four parameters for determining the operating condition of the massage chair 1, namely, strength, pulse, region, and simultaneous leg, and the massage operation is performed by a combination of these four parameters. There are three levels of strength: strong, medium, and weak.
There are three stages, fast and slow, and the parts are neck / shoulder, back,
There are six types of back, waist, buttocks, and thighs, and simultaneous legs have two stages, ON and OFF.

The Ag-AgCl electrode 4 measures the brain wave of the user 2 which is one of the physiological indices, and is made of a hair band-shaped member so that it can be easily worn. It is configured to detect the brain waves of the user 2 who is wearing the head and performing the massage operation. This electrode 4
Has three electroencephalogram measurement electrodes, one located at the center of the forehead, and the other two located at both sides thereof. In addition, as physiological indices measured in the engineering field, there are electroencephalographic activity, blood pressure, heart rate, respiration, pulse wave, etc., in addition to electroencephalogram, but the electroencephalogram is the most reliable index from stability,
Therefore, EEG is used.

Although the derivation of the electroencephalogram may be performed by monopolar induction for detecting a difference from a reference electrode such as an earlobe, the detection is performed by bipolar induction here. The hyperbolic induction is, for example, a one-channel electroencephalograph derived from the forehead bipolar based on the forehead, as described in Japanese Patent Application Laid-Open No. 10-262942. Detects a minute potential by deriving the bipolar.

The signal processing unit 5 includes the Ag-AgCl electrode 4
After amplifying the brain waves measured in step (1) and performing A / D conversion, the waveform information of the brain waves is transmitted to the computer 6 as digital data. Here, the signal processing unit 5
Exchange data from computer to computer 6 via RS232
This is done using a wire such as C. Note that this data exchange can also be performed by a wireless method using radio waves or infrared rays. When the wireless method is used, the user 2 of the massage chair 1 can hold the signal processing unit 5 or wear it on a belt or the like, which allows the user 2 to freely move around and use the user 2 in the electroencephalogram measurement. The burden can be reduced.

The computer 6 includes a massage effect estimating means for estimating a massage effect on the user 2 from the brain waves of the user 2, and a computer for estimating the massage effect estimated by the massage effect estimating means. The learning unit includes learning means for learning a massage pattern by increasing or decreasing the reward strength used for reinforcement learning for each parameter state, and massage control means for performing massage control using the parameters learned by the learning means.

FIG. 2 is a block diagram showing the overall control configuration. Of the three electrodes of the Ag-AgCl electrode 4, two electrodes detected between the left and right electrodes with reference to the electrode at the center are shown. After the voltage is operated and amplified by the operational amplifier 51 of the signal processing unit 5 to obtain a one-channel brain wave signal, the brain wave signal is further amplified to a level at which the A / D conversion can be performed. They are converted to digital signals.

The brain wave signal converted to a digital signal output from the signal processing unit 5 is taken into a computer 6. The computer 6 estimates a massage effect based on the brain wave signal, and optimizes a massage pattern based on the estimation result.

FIG. 3 is a flowchart showing the overall control procedure. In step S1, the user 2 uses the Ag-AgCl electrode 4 to perform the control.
Physiological indices are detected from the brain waves of the subject, and in step S2, the massage effect is estimated by the computer 6, and in step S2
In 3, the computer 6 learns to obtain an optimal massage pattern for the user based on the estimation result of the massage effect. In step S4, the computer 6 controls the massage chair 1 based on the learning result.

Next, software processing in the computer 6 will be described. The digital signal of the electroencephalogram is time-series data of the potential sampled at regular intervals. First, this is divided into regular intervals, and a fast Fourier transform (hereinafter, referred to as FFT) function 61 performs fast Fourier transform processing. The power spectrum PS is calculated. FFT
In the processing, since the number of data needs to be a power of 2, for example, when the sampling frequency is 100 Hz and the number of data is 512 (2 9), 4.27.
A second is defined as one section. In the FFT process, it is preferable to use a Hanning window to reduce the influence of the cutting error.

The power spectrum thus obtained is input to the neural network 62 to estimate comfort. This neural network 62 is learned by using data in which the correspondence between the power spectrum and the user's mental state is known in advance. Is output.

Specifically, first, a plurality of brain wave power spectra when the user is judged comfortable by the subjective evaluation of the user and a plurality of brain wave power spectra when the user feels uncomfortable are prepared. The massage machine 1 is actually used for a neural network trained using a general learning algorithm such as an error backpropagation method so that 1 is output when a spectrum when it is not comfortable is input. By inputting the brain wave power spectrum of user 2 of
The mental state at that time is calculated as a numerical value of 0 to 1. In this case, 0 indicates the least comfortable state and 1 indicates the most comfortable state.

The neural network may learn using the brain waves of the user who actually uses the apparatus, or may learn using brain waves collected in advance for a plurality of subjects.
In the latter case, there is an advantage that the user can use the massage system immediately, and in the former case, there is an advantage that the psychological state of the user can be more accurately estimated.

An optimal massage pattern is obtained from the comfort value thus obtained. Next, a method of optimizing the massage pattern will be described. The optimization of the massage pattern is performed by the reinforcement learning function 63 by a reinforcement learning method in which a lottery using random numbers is used as an action determination method. FIG. 4 is a diagram showing a state of a probability distribution at the time of random number drawing at a stage where learning has progressed to some extent, and shows four graphs corresponding to parameters of a massage pattern.

For example, among the parameters, there are two states where the strength is strong and weak, and the area P1 of the oblique hatching rising to the right represents the establishment when the strength is strong, and the area P2 of the oblique hatching rising to the left is It indicates the establishment when the strength is weak. The operating condition determining function 64 determines which state is to be selected by a lottery using random numbers for parameters having such a probability distribution. The positions of the white circles in each graph correspond to numerical values assigned by random numbers.

Similarly, a lottery is performed for another parameter, “pulse”, “leg simultaneous”, and “part”, and a pattern is selected based on the assigned numerical value. Note that the graph of “pulse” corresponds to three states of “none”, “slow”, and “fast” from the left, and the graph of “leg simultaneous” includes two states of “ON” and “OFF” from the left. Corresponding to the state, the graph of “region” corresponds to six states of “neck / shoulder”, “back”, “back muscle”, “hip”, “butt”, and “peach” from the left. The graph of FIG. 4 shows that the strength is “weak”, the pulse is “fast”, and the “leg simultaneous” is “O”.
"N" and "ass" are selected as the "parts".

The massage is actually performed with the massage pattern selected in this manner, and the probability distribution of the next lottery is determined according to how the psychological state (the degree of comfort) of the user changes according to the result. Change. By repeating this, the establishment of the massage pattern that the user feels most comfortable gradually increases, and optimization can be realized.

The change of the probability distribution in this optimization is determined by the procedure shown in the following equation. The amount of increase δ (t) of the reward at time t for each state j of each parameter i is

[0031]

(Equation 1)

Here, S (t) is a numerical value representing the degree of comfort obtained by the above-described neural network 62, k and u are natural numbers, and u represents the number of unit evaluations.
(One evaluation time is 100 Hz sampling frequency,
4.27 seconds when the number of data is 512), and each state j
Xij (t) = Xij (t-1) + δ (t) · Wij (2) where Wij is a weighting factor and is appropriately set according to the number of states for each parameter. . Is defined. Further, a reward strength Rij (t) is calculated from Xij (t) by the following equation.

[0033]

(Equation 2)

The probability Pij of each state is determined by the reward intensity ratio. That is, the occurrence probability Pij is

[0035]

(Equation 3)

Is as follows.

The calculation for estimating the comfort by the neural network 62 can be realized substantially in real time. Therefore, as shown in the above equation (1), the probability at the time of the lottery is increased or decreased by a change in comfort, that is, when the comfort is increased, the probability of the state is increased, and when the comfort is reduced, By repeating the operation of performing a lottery with a reduced probability, performing a massage of the assigned massage pattern, and evaluating the increase or decrease of comfort again, the probability of only the massage pattern that the user feels most comfortable increases, and the massage You will focus on patterns.

When the massage pattern is determined by such a method, for example, when the user feels that both the waist and the body are comfortable, in principle, not only one is selected, but both are selected. Have the same probability of occurrence, and both will massage at approximately the same rate.
That is, the control signal is transmitted from the control signal transmitting means 65 constituting the massage control means to the massage chair 1 based on the optimal massage pattern determined in this way, and the massage chair 1 is driven.

Also, once the massage pattern converges to the most comfortable one, the comfort may decrease due to the user's getting used to or getting tired. In this case, the occurrence probability of the converged massage pattern starts to decrease, and another Massage will be performed in a massage pattern, and a comfortable massage can always be performed.

Furthermore, since the probability of occurrence of another massage pattern does not necessarily become 0 after having converged to the optimal massage pattern, massage is occasionally performed with a massage pattern other than the optimal massage pattern. This produces an effect that the user is less likely to get tired and used.

In addition, the action determination method can simply have a high reward strength without using a lottery based on random numbers. That is, the state in which Rij calculated by the above equation (3) is the maximum is checked, and the state corresponding to the parameter is selected for each parameter. Just choose. In this case, although there is a problem of getting tired and used, massage can be performed in a pattern that is always considered to be the most comfortable.

Furthermore, the subjective evaluation result of the user's mental state can be input to the computer 6 using the joystick 3. In order to convert the user's subjective amount into a numerical value using the joystick 3, the user is instructed in advance to change the amount of tilting the joystick 3 according to the degree of comfort, and the amount of actual tilting is determined. What is necessary is just to detect, sequentially digitize, and take the average value. Also, for example, only when you feel comfortable,
May be instructed to tilt more than a predetermined value, and whether the joystick 3 is tilted beyond the predetermined value is detected at an appropriate cycle and averaged over an appropriate time. In these methods, if the average time is the same as one unit time for estimating the degree of satisfaction from brain waves, for example, 4.27 seconds, data processing is easy.

Using the numerical values of the subjective evaluation results obtained in this way, reinforcement learning can be performed by using equations equivalent to the above equations (1) to (4), and an optimal massage pattern can be obtained. Can be determined. As described above, the reinforcement learning can be performed by using both the subjective evaluation result and the degree of satisfaction by the electroencephalogram, or by using only one of them.

In the above-described embodiment, when the internal state quantity is changed by the equation (2), the weighting coefficient Wij multiplied by the reward increase amount δ (t) is appropriately set according to the number of states for each parameter. This means, for example, that the state of the parameter “part” has six necks / shoulders, back, back muscles, hips, buttocks,
There are only two states of the parameter “strength”, strong and weak. Thus, the amount of change is given to those having different numbers of states with the same weight. This means that, for example, the weight of the parameter “part” is set to １ ／ of the weight of the parameter “strength”.

By the way, the characteristics of human brain waves vary greatly from person to person, and individual differences in the degree of satisfaction estimated based on these brain waves are also large. For example, some people change in a place where the numerical value of the degree of satisfaction is high, and others change the value of the degree of satisfaction with a low value even if the mental state changes in the same manner. In addition, there are some people who have relatively large variations in the degree of satisfaction in the measurement under the same conditions, and some people who have a small variation. Also, a person with a large variation in the degree of satisfaction does not always have a large change in the psychological state.

However, since the amount of increase in reward is defined by a change in satisfaction as shown in the above equation (1), a person with a large variation in satisfaction rapidly changes the reward amount, and conversely The reward amount does not change much for a person with a small value, and the learning speed of reinforcement learning may differ.

Therefore, the variation in the degree of satisfaction is added to the weight coefficient. Specifically, a weight determined by integrating the reciprocal of the standard deviation of the degree of satisfaction measured in a resting state before starting the reinforcement learning by operating the massage chair 1 with respect to the weight determined from the number of parameter states, etc. Wij.

By using such a weighting coefficient Wij, the reinforcement learning proceeds at substantially the same speed for both a person with a large variation in satisfaction and a person with a small degree of satisfaction, and an optimal massage pattern can be accurately determined.

Here, the standard deviation of the satisfaction level is a value at the time of rest, but a value obtained by recalculating the satisfaction level at the stage of performing the reinforcement learning, as needed, may be used. In this case, even if the number of satisfaction samples is increased without limit and recalculated,
When a certain number is reached, old ones may be discarded and the same number of samples may always be calculated.

In the above-described embodiment, the brain wave is used as a physiological index, and the degree of satisfaction estimated from the brain wave is used. However, the present invention is not limited to this. Satisfaction levels estimated using other physiological indices such as heart rate, blood pressure, and skin electrical response may be used.

Further, a storage unit for storing the reward strength is provided,
If the reward intensity at the time when the user last receives the massage is stored in this storage unit, the user can receive the massage that is suitable for himself from the beginning next time.
In this case, if a storage unit corresponding to a plurality of users is provided, and the reward intensity is individually stored for each user, all the members can be designated by specifying which storage pattern to use before receiving the massage. You can receive a massage that suits you from the start.

[0052]

As described above in detail, according to the present invention, the massage operation can be performed with the optimal massage pattern, and even when the place where the user feels comfortable changes, the massage pattern is changed in real time to optimize the massage pattern. Massage operation can be performed with a simple massage pattern.

According to the second aspect of the present invention, the ratio of the reward strength for each parameter state is set as a probability in learning, and the state of each parameter is determined by lottery using random numbers. Massage can be performed with an unexpected massage pattern, and the tiredness caused by being repeatedly massaged in the same place or being massaged in a predetermined pattern can be prevented.

According to the third aspect of the present invention, a value for increasing / decreasing the learning reward strength is further integrated with the estimated increase / decrease amount of the massage effect by the reciprocal of the variation in the evaluation result. Since the setting is made, even if there is a correlation between a change in emotion such as pleasure or discomfort and a change in the physiological index associated therewith, or even if there are individual differences in the characteristics of the physiological index itself, accurate reinforcement learning can be performed at the same learning speed.

[Brief description of the drawings]

FIG. 1 is a diagram showing an overall configuration of a massage machine according to an embodiment of the present invention.

FIG. 2 is a block diagram showing a control configuration of the massage machine in the embodiment.

FIG. 3 is a flowchart showing an overall control procedure of the massage machine in the embodiment.

FIG. 4 is a diagram showing a state of a probability distribution at the time of random number drawing at a stage where learning has progressed to some extent in the embodiment.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 ... Massage chair 4 ... Ag-AgCl electrode 5 ... Signal processing part 6 ... Computer 62 ... Neural network 63 ... Reinforcement learning function

──────────────────────────────────────────────────の Continued on the front page (51) Int.Cl. ⁷ Identification symbol FI Theme coat ゛ (Reference) A61H 15/00 350 A61H 23/02 390 23/02 390 A61B 5/04 322 (72) Inventor Mika Watanabe Shizuoka 6-78, Minami-cho, Mishima-shi, Toshiba Tec Corporation Mishima Office (72) Inventor Hiroshi Nakazawa 1-16-26 Matsugaoka, Tokorozawa-shi, Saitama F-term (reference) 4C027 AA03 CC00 EE05 FF01 GG11 GG15 JJ00 JJ03 KK00 KK03 4C074 AA03 AA04 AA05 BB05 CC01 CC13 CC17 GG02 GG03 GG05 4C100 AD00 AE00 BA03 BB02 BC11 CA03 CA13 CA15 DA04 DA05 DA06 DA10 EA10

Claims (3)

[Claims] In a massage machine having a plurality of parameters for determining operating conditions, an effect of massage for a user when used is estimated from a physiological index of the user and / or evaluated based on evaluation information input from the user. Massage effect estimating means for estimating the massage pattern based on the massage pattern by increasing or decreasing the reward intensity used for reinforcement learning for each state of each of the parameters in response to an increase or decrease in the evaluation result of the massage effect estimated by the massage effect estimating means. And a massage control means for performing massage control using parameters learned by the learning means. 2. The massage machine according to claim 1, wherein the learning in the learning means determines the ratio of the reward strength for each parameter state as a probability, and determines the state of each parameter by random drawing. 3. A value for increasing or decreasing the reward strength of learning in the learning means is obtained by integrating at least the reciprocal of the variation of the evaluation result with the increase / decrease amount of the evaluation result of the massage effect estimated by the massage effect estimating means. The massage machine according to claim 1 or 2, wherein: