JP6297951B2 - Motion estimation device - Google Patents

Description

  The present invention relates to a motion estimation device and a motion estimation method, and more particularly to a motion estimation device and a motion estimation method for estimating the motion of a patient who is performing rehabilitation.

  A person whose motor function is impaired can recover the motor function by performing rehabilitation. In recent years, various devices for appropriately performing such rehabilitation have been developed.

  Patent Document 1 discloses a technique related to a motion assisting device for assisting human motion. In the motion assisting device disclosed in Patent Document 1, a dynamic identification parameter specific to the wearer is identified by the parameter identification unit in a state worn by the wearer, and the control device is based on an equation of motion substituted with the identified dynamic parameter. To control the drive source. Thereby, the movement assistance apparatus which is not influenced by fluctuation factors, such as a wearer's individual difference and a physical condition, can be provided.

JP 2006-204426 A

  In the technique disclosed in Patent Document 1, a wearer-specific dynamic parameter is identified by a parameter identification unit in order to compensate for individual differences among wearers, and a drive source is based on a motion equation into which the identified dynamic parameter is substituted. Is controlling.

  However, in the technique disclosed in Patent Document 1, the wearer-specific dynamic parameters are only used for gravity compensation and inertia compensation, and are not used for the control for estimating the wearer's intention. For this reason, the technique concerning patent document 1 cannot be applied to the control system which estimates a patient's intention continuously and assists like rehabilitation.

  That is, in rehabilitation, it is required to continuously estimate the patient's intention. Therefore, there is a need for a motion estimation device and a motion estimation method that reads a patient's intention from a patient's biological signal and estimates the patient's motion in real time based on the biological signal.

  The motion estimation device according to the present invention includes an operation unit that is operated by a patient who is performing rehabilitation, an operation amount detection unit that detects an operation amount of the operation unit, and the patient when the patient operates the operation unit. A biological signal detection unit that detects a biological signal of the patient, and a motion estimation unit that estimates the motion of the patient using the operation amount detected by the operation amount detection unit and the biological signal detected by the biological signal detection unit And the motion estimation unit calculates a true value of the patient's exercise amount calculated using the operation amount of the operation unit and a biological signal of the patient when the patient operates the operation unit. A weight vector generation unit that generates a sparse weight vector using the sparse weight vector generated by the weight vector generation unit and the biological signal detection unit when the patient is performing rehabilitation. And a motion estimation value generating unit for generating a motion estimation value of the patient using a biological signal detected in real time.

  The motion estimation method according to the present invention uses the operation amount of an operation unit operated by a patient undergoing rehabilitation and the patient's biological signal when the patient operates the operation unit. A motion estimation method for estimating the sparseness using a true value of the patient's motion amount calculated using the operation amount of the operation unit and a biological signal of the patient when the patient operates the operation unit. The patient using a weight vector generation step for generating a weight vector, the sparse weight vector generated in the weight vector generation step, and a biological signal detected in real time when the patient is performing rehabilitation A motion estimation value generating step of generating a motion estimation value of

  In the present invention, a sparse weight vector is generated in advance using the true value of the patient's momentum and the patient's biological signal. And the patient's motion estimated value is produced | generated using the biometric signal detected in real time when the patient is performing rehabilitation, and this sparse weight vector. At this time, in the present invention, since the sparse weight vector is used when generating the motion estimation value of the patient, the calculation load when calculating the motion estimation value can be reduced. Therefore, even when a biological signal is continuously detected as in rehabilitation, a motion estimation value can be calculated quickly, and a motion estimation value based on the patient's biological signal can be obtained in real time. .

  According to the present invention, it is possible to provide a motion estimation device and a motion estimation method capable of estimating a patient's motion in real time.

It is a block diagram which shows the motion estimation apparatus concerning embodiment. It is a figure for demonstrating the state in which the patient is using the motion estimation apparatus. It is a side view which shows the robot for rehabilitation. It is a block diagram for demonstrating a motion estimation part in detail. It is a flowchart for demonstrating operation | movement of the motion estimation apparatus concerning embodiment. It is a graph which shows the result (result of speed estimation) which estimated the motion of the patient using the motion estimation apparatus concerning embodiment (use the data of a learning phase). It is a graph which shows the result (position estimation result) which estimated the motion of the patient using the motion estimation apparatus concerning embodiment (use the data of a learning phase). It is a graph which shows the result (result of speed estimation) which estimated the motion of the patient using the motion estimation apparatus concerning embodiment (use the data of a test phase). It is a graph which shows the result (position estimation result) which estimated the motion of the patient using the motion estimation apparatus concerning embodiment (use the data of a test phase). It is a table | surface which shows the performance comparison of the motion estimation apparatus concerning embodiment (use the data of a learning phase). It is a table | surface which shows the performance comparison of the motion estimation apparatus concerning embodiment (using the data of a test phase).

Embodiments of the present invention will be described below with reference to the drawings.
FIG. 1 is a block diagram showing a motion estimation apparatus 1 according to the present embodiment. As shown in FIG. 1, the motion estimation apparatus 1 according to the present embodiment includes a rehabilitation robot 10, a biological signal detection unit 14, a control unit 15, a servo amplifier 17, and a display unit 18. The rehabilitation robot 10 includes an operation unit 11, a motor 12, and an operation amount detection unit 13. The control unit 15 includes a motion estimation unit 16.

  Since the present invention is characterized in that the motion of the patient is estimated, the apparatus shown in FIG. 1 will be described as a motion estimation apparatus. However, the motion estimation apparatus 1 can also be used as a rehabilitation apparatus. When used as the motion estimation device 1, at least the operation unit 11, the operation amount detection unit 13, the biological signal detection unit 14, and the motion estimation unit 16 may be provided as constituent elements.

  The operation unit 11 included in the rehabilitation robot 10 is used by the patient 19 to perform rehabilitation. The motor 12 is configured using a servo motor, for example, and applies torque to the operation unit 11. The operation amount detection unit 13 detects the operation amount of the operation unit 11. For example, the operation amount detection unit 13 can be configured using an encoder (hereinafter also referred to as the encoder 13).

  As shown in FIGS. 2 and 3, the rehabilitation robot 10 includes a casing 31, joints 32 and 34, arms 33 and 35, and an operation unit 11. The arm 33 is attached to the housing 31 via a joint 32. That is, the arm 33 is rotatably attached to the housing 31. The arm 33 and the arm 35 are rotatably connected via a joint 34. The operation unit 11 is rotatably attached to the tip of the arm 35.

  A motor 12_1 for applying torque to the arm 33 is attached to the joint 32. The motor 12_1 is driven in accordance with the drive signal supplied from the servo amplifier 17. In addition, an encoder 13_1 for detecting the rotation angle of the arm 33 with respect to the casing 31 is attached to the joint 32. The rotation angle information 21_1 output from the encoder 13_1 is supplied to the control unit 15.

  A motor 12_2 for applying torque to the arms 33 and 35 is attached to the joint 34. The motor 12_2 is driven in accordance with the drive signal supplied from the servo amplifier 17. Further, an encoder 13_2 for detecting the rotation angle of the arm 35 with respect to the arm 33 is attached to the joint 34. The angle information 21_2 output from the encoder 13_2 is supplied to the control unit 15.

Thus, the operation unit 11 is configured to be displaceable on the two-dimensional plane (p _x , p _y ) by the upper limb motion of the patient 19 (see FIG. 2). The patient 19 operates the operation unit 11 in accordance with the instruction displayed on the display unit 18. The operation amount when the patient 19 moves the operation unit 11 is supplied to the control unit 15 from the encoders 13_1 and 13_2 as angle information 21_1 and 21_2.

  The biological signal detection unit 14 illustrated in FIG. 1 detects a biological signal of the patient 19 when the patient 19 operates the operation unit 11. As the biological signal detection unit 14, a myoelectric potential sensor that detects the myoelectric potential of the exercise site of the patient 19 can be used. Further, an electroencephalogram sensor that detects an electroencephalogram (EEG) of the patient 19 and a cortical electroencephalogram sensor that detects an electrocorticogram (EcoG) of the patient 19 may be used. Moreover, you may use combining a myoelectric potential sensor, an electroencephalogram sensor, and a cortical electroencephalogram sensor. Below, the case where a myoelectric potential sensor is used as the biological signal detection part 14 is demonstrated.

  When a myoelectric potential sensor is used as the biological signal detection unit 14, for example, as shown in FIG. 2, the myoelectric potential sensor 14_1 is disposed at a position corresponding to the greater pectoral muscle (PMJC) of the patient 19, and the rear part of the deltoid muscle (DELS). The myoelectric potential sensor 14_2 is disposed at a position corresponding to (). Here, the great pectoral muscle and the back of the deltoid muscle are muscles that antagonize each other. Further, the myoelectric potential sensor 14_3 is arranged at a position corresponding to the long biceps brachii (BILH) of the patient 19, and the myoelectric sensor 14_4 is arranged at a position corresponding to the triceps outer head (TRIA). Here, the biceps long head and the triceps lateral head are muscles that antagonize each other.

The myoelectric potential signals emg ⁱ _k detected by the myoelectric potential sensors 14_1 to 14_4 (hereinafter collectively referred to as the myoelectric potential sensor 14) are supplied to the control unit 15.

The motion estimation unit 16 included in the control unit 15 includes an operation amount 21 of the operation unit 11 detected by the encoder 13 and a biological signal 22 (myoelectric signal emg ⁱ _k ) detected by the biological signal detection unit (myoelectric potential sensor) 14. And the motion of the patient 19 is estimated.

  Further, the control unit 15 controls the display unit 18 so that the direction in which the patient 19 should operate the operation unit 11 is displayed. When used as a rehabilitation device, the control unit 15 outputs a control target value to the servo amplifier 17 so that the motor 12 generates torque according to the motion estimation value generated by the motion estimation unit 16.

  FIG. 4 is a block diagram for explaining the motion estimation unit 16 in detail. As shown in FIG. 4, the motion estimation unit 16 includes a weight vector generation unit 40 and a motion estimation value generation unit 50. The weight vector generation unit 40 includes a preprocessing unit 41, a time shift operation unit 42, a forward kinematics processing unit 43, and a weight vector calculation unit 44. The motion estimated value generation unit 50 includes a preprocessing unit 51, a time shift operation unit 52, and a motion estimated value calculation unit 53.

The weight vector generation unit 40 uses the true value y of the amount of movement of the patient 19 calculated using the operation amount (angle information) detected by the encoder 13 and the myoelectric potential signal emg ⁱ _k detected by the myoelectric potential sensor 14. To generate a sparse weight vector θ _h . The motion estimation value generation unit 50 uses the sparse weight vector θ _h generated by the weight vector generation unit 40 and the myoelectric potential signal emg ⁱ _k detected by the myoelectric potential sensor 14 when the patient 19 is performing rehabilitation. The motion estimate value y _h of the patient 19 is generated.

  Here, the sparse weight vector is a vector with a small number of non-zero elements. In this embodiment, the sparse weight vector is a sparse weight so that the number of non-zero elements included in the weight vector is equal to or less than a predetermined number. Generate a weight vector.

Hereinafter, the process of the motion estimation unit 16 will be described in detail using the block diagram of the motion estimation unit 16 shown in FIG. 4 and the flowchart shown in FIG. In the flowchart shown in FIG. 5, the processing of steps S1 to S5 for generating the sparse weight vector θ _h is a learning phase (weight vector generation step: corresponding to the processing of the weight vector generation unit 40), and motion estimation The processing of steps S6 to S8 for generating the value y _h is a test phase (motion estimation value generation step: corresponding to the processing of the motion estimation value generation unit 50).

First, processing (processing in the learning phase) of the weight vector generation unit 40 will be described. The weight vector generation unit 40 receives the myoelectric potential signal emg ⁱ _k (step S1). Here, the myoelectric potential signal emg ⁱ _k is a signal measured by the 4-channel myoelectric potential sensors 14_1 to 14_4 shown in FIG. Next, the pre-processing and time shift operations on electromyogram ^emg _{i k} (Step S2).

Specifically, the preprocessing section 41, after performing rectification processing on the myoelectric potential signals emg ⁱ _k of 4 channels, thereby emphasizing the low frequency components of the myoelectric potential signal over time filter. At this time, a 5th-order band pass filter G _BPF (z) is applied to each channel as a time filter to pass a band from 0.5 Hz to 2 Hz. The processing at this time is shown in Expression (1).

Here, k is the time, i is the channel number of the myoelectric potential sensors 14_1 to 14_4, and i = 1, 2, 3, and 4 are the greater pectoral muscle (PMJC), deltoid muscle (DELS), and biceps. It corresponds to the long muscle head (BILH) and the triceps lateral head (TRIA). Further, | egg ⁱ _k | represents a rectification process. z indicates z conversion.

Then, the preprocessing unit 41 performs differential processing in the time direction shown in equation (2), to create a myoelectric potential (EMG) speed signal d ⁱ _k related to low-frequency components. By applying Expression (2), it is possible to improve the responsiveness at the time of decoding.

Then, the time shift operation unit 42 performs the time shift operation for each channel with respect to myoelectric potential (EMG) speed signal ^d _{i k} obtained by the above formula (2), to create a response matrix _{A meas.} Below, Formula (3) showing time shift operation is shown.

  Next, the learning phase processing unit 40 inputs the operation amount (angle information) detected by the encoder 13 (step S3). Thereafter, the true value y of the momentum of the patient 19 is calculated by forward kinematics processing using the operation amount (angle information) detected by the encoder 13 (step S4).

  Specifically, the forward kinematics processing unit 43 calculates the true value y of the momentum of the patient 19 by forward kinematics processing using the operation amount (angle information) detected by the encoder 13. The true value y of the momentum at this time is shown in Formula (4).

Here, M is the number of measured and calculated samples, and the sampling frequency at that time is 128 Hz. Equation (4) shows a case where the hand speed v _x in the x direction of the patient 19 is estimated. Y-direction of the hand velocity v _y of the patient _19, the hand displacement p _x in the x _direction, the y-direction of the hand displacement p _y can be obtained by changing the appropriate motion physical value the true value of the equation (4) .

Next, the weight vector calculation unit 44 generates a sparse weight vector θ _h using the response matrix A _meas and the true value y of the momentum (step S5). Hereinafter, a method for generating the supersed weight vector θ _h will be specifically described.

A linear decoding model based on the electromyogram (EMG) velocity signal d ⁱ _k can be expressed by the following equation (5).

Here, C is the number of channels of the myoelectric sensors 14_1 to 14_4 (C = 4), N is the number of time shifts used in decoding (N = 14), and a and θ ⁱ _j are model weighting factors. When Expression (5) is expressed using a matrix, it can be expressed as Expression (6) below.

Here, A _meas is a response matrix, and θ is a weight vector. In the above equation (6), since the true value y of the momentum and the response matrix A _meas are known, the weight vector θ can be obtained using the above equation (6). At this time, the sparse coding is used when obtaining the weight vector θ, so that the sparse weight vector θ _{h in which} the number of non-zero elements is smaller than the weight vector θ (hereinafter also referred to as an estimated value θ _h of the weight vector). ) Can be generated. Hereinafter, a case where the estimated value θ _h of the weight vector is generated using sparse coding will be described in detail.

The problem of obtaining an estimated value θ _h of the weight vector using sparse coding can be expressed as the following equation (7).

Here, || θ || ₀ is the number of non-zero elements in the weight vector θ, and Equation (7) is very sparse while satisfying the condition, that is, obtains a weight vector composed of a small number of non-zero elements. Represents that.

  In order to more accurately represent the condition part of Equation (7), the problem formulation is rewritten as shown in Equation (8) below.

  Here, ε is the maximum allowable error. Then, the target value of sparse expression, that is, the target value S of the number of non-zero elements is given as the following expression (9), and the expression (8) is numerically calculated by a recursive algorithm. For this calculation, for example, an OMP (Orthogonal Matching Pursuit) method having excellent convergence of recursive calculation can be used.

By such processing, the weight vector generating unit 44 can generate a sparse weight vector theta _h. In the above example, sparse coding by the OMP method is used to obtain a sparse weight vector. However, since the final object of the present invention is to obtain sparse weights for the linear decoding model, another method of sparse coding may be used. Furthermore, some of the weight elements obtained by the least square method may be set to zero to obtain an optimum weight vector that satisfies Expression (7) (see Expression (11)).

Next, processing (processing in the test phase) of the motion estimated value generation unit 50 will be described. When the patient 19 operates the operation unit 11, the myoelectric potential signal ^emg _{i k} is output in real time from the myoelectric potential sensor 14_1～14_4 of four channels shown in FIG. The estimated motion value generation unit 50 inputs the myoelectric potential signal emg ⁱ _k (step S6). Next, the pre-processing and time shift operations on electromyogram ^emg _{i k} (step S7).

Specifically, the preprocessing section 51, 4 after performing rectification processing on the myoelectric potential signals emg ⁱ _k channels, to emphasize the low frequency components of the myoelectric potential signal over time filter. Thereafter, the pre-processing unit 51 performs time direction difference processing, to create a myoelectric potential (EMG) speed signal d ⁱ _k related to low-frequency components. Note that the processing in the preprocessing unit 51 is the same as the processing in the preprocessing unit 41 described above, and thus a duplicate description is omitted.

Then, the time shift operation unit 52 performs the time shift operation for each channel with respect to myoelectric potential (EMG) speed signal ^d _{i k} obtained above, to create a response matrix _{A meas.} Note that the processing in the time shift operation unit 52 is the same as the processing in the time shift operation unit 42 described above, and thus a duplicate description is omitted.

Next, the motion estimation value calculation unit 53 uses the response matrix A _meas generated by the time shift operation unit 52 and the sparse weight vector θ _h generated by the weight vector generation unit 40 to estimate the motion of the patient 19. A value y _h is generated (step S8).

Specifically, the motion estimation value calculation unit 53 sets the response matrix A _meas generated by the time shift operation unit 52 and the sparse weight vector θ _h generated by the weight vector generation unit 40 to the following equation (10). And the motion estimation value y _h of the patient 19 is calculated.

As described above, the motion estimation value generation unit 50 (test phase) takes into account the time-invariant linear decoding model for motion estimation (decoding) and performs preprocessing (rectification) on the measured myoelectric potential signal emg ⁱ _k . By inputting the response matrix A _meas subjected to the time shift operation, the motion estimation value y _h (hand speed, displacement) is output. In the test phase, the sparse weight vector θ _h is used as a time-invariant parameter.

When the motion estimation device 1 is used as a rehabilitation device, the rehabilitation robot 10 is controlled using the motion estimation value y _h generated by the motion estimation value generation unit 50 (step S9). Specifically, the control unit 15 shown in FIG. 1, the motor 12 is to generate a torque, and outputs the control target value to the servo amplifier 17 in accordance with the generated motion estimation value y _h. By such control, assist control according to the intention of the patient 19 can be performed.

Thereafter, by repeating the processing of the test phase shown in FIG. 5 (steps S6 to S9), the motion of the patient can be estimated in real time, and the assist control according to the intention of the patient 19 can be continuously performed. . Also, if the need to newly generate a sparse weight vector theta _h occurs, it is implementing the processing of the learning phase (steps S1-S5).

  Next, the performance comparison of the motion estimation apparatus according to the present embodiment will be described. For performance comparison, three types of decoders (the weight vector calculation unit 44, hereinafter also referred to as a decoder) were designed. The decoder A is a decoder in which the target value S of the non-zero element in Expression (9) is set to S = 21. The decoder B is a decoder in which the target value S of the non-zero element of Equation (9) is set to S = 9. The decoder C is a decoder designed by the least square method.

When the estimated value θ _h of the weight vector is obtained using the least square method, the true value y of the motion output from the forward kinematics processing unit 43 and the estimated motion value y _h obtained by the above equation (10) are used. The weight that minimizes the square error is calculated using the following equation (11).

  The number of elements of the weight vector designed in all decoders A to C is 57 dimensions. In decoder A, 21 of them are non-zero elements, and in decoder B, 9 of them are non-zero elements. The other elements are zero elements.

FIG. 6 and FIG. 7 are diagrams showing the results of estimating the patient's motion using the decoder A, and each showing the velocity and position obtained using the data in the learning phase. The upper graph in FIG. 6 shows the velocity v _{x in the} x direction, and the lower graph shows the velocity vy in the _y direction. The upper graph in Figure 7 position p _x in the x _direction, the lower graph shows the position p _y in the y direction. In the figure, the solid line indicates the motion estimation value y _h obtained using the sparse weight vector θ _h , and the broken line indicates the motion true value y obtained from the angle information of the encoder.

Here, the result obtained using the data of the learning phase is the motion estimation value y using the same myoelectric signal emg ⁱ _k as the myoelectric signal emg ⁱ _k used when obtaining the sparse weight vector θ _h. It means that _h is obtained. That means the case where the myoelectric potential signals emg ⁱ _k supplied to the electromyogram emg ⁱ _k preprocessing section 51 is supplied to the preprocessing unit 41 of FIG. 4 of the same.

As shown in FIG. 6, when estimating the motion of the patient using the decoder A, the speed v _y of the velocity v _x and y direction of the x-direction, the motion estimation value y _h exercise true value y is highly correlated Indicated. Further, as shown in FIG. 7, when estimating the motion of the patient using the decoder A, the position p _y position p _x and y direction of the x-direction, movement true value y and motion estimation value y _h high correlation The relationship was shown.

FIG. 8 and FIG. 9 are diagrams showing the results of estimating the patient's motion using the decoder A, and each showing the velocity and position obtained using the data of the test phase. The upper graph in FIG. 8 shows the velocity v _{x in the} x direction, and the lower graph shows the velocity vy in the _y direction. The upper graph of Figure 9 is the position p _x in the x _direction, the lower graph shows the position p _y in the y direction. In the figure, the solid line indicates the motion estimation value y _h obtained using the sparse weight vector θ _h , and the broken line indicates the motion true value y obtained from the angle information of the encoder.

Here, the results obtained by using the data of the test phase, the motion estimation value by using a myoelectric potential signal emg ⁱ _k different electromyogram emg ⁱ _k used in obtaining the sparse weight vector theta _h y _h Means that That means the case where the myoelectric potential signals emg ⁱ _k supplied to the electromyogram emg ⁱ _k preprocessing section 51 is supplied to the preprocessing unit 41 of FIG. 4 differs.

As shown in FIG. 8, when estimating the motion of the patient using the decoder A, the speed v _y of the velocity v _x and y direction of the x-direction, the motion estimation value y _h exercise true value y is highly correlated Indicated. Further, as shown in FIG. 9, when estimating the motion of the patient using the decoder A, the position p _y position p _x and y direction of the x-direction, movement true value y and motion estimation value y _h high correlation The relationship was shown.

In order to quantitatively evaluate these results, the correlation coefficient R between the motion true value y and the motion estimated value y _h shown in FIGS. 6 to 9 was obtained. Further, in the case of using the decoders B and C, the correlation coefficient R between the motion true value y and the motion estimated value y _h was obtained. FIGS. 10 and 11 show the correlation coefficient R between the motion true value y and the motion estimated value y _h when the decoders A to C are used. FIG. 10 shows a performance comparison when the learning phase data is used, and FIG. 11 shows a performance comparison when the test phase data is used.

In the results shown in FIG. 10, the correlation coefficient is highest when the decoder C is used, and then the correlation coefficient decreases in the order of the decoder A and the decoder B. This tendency was substantially the same in the results shown in FIG. The reason why the correlation coefficient is the highest in the decoder C seems to be that the decoder C designed by the least square method _applies a 57-dimensional weight vector to all elements of the response matrix A _meas . On the other hand, since decoders A and B designed by sparse coding only apply 21-dimensional and 9-dimensional sparse weights, the information for decoding is reduced and the correlation coefficient is reduced. Seem.

  However, paying attention to each correlation coefficient, both the decoders A and B show a strong correlation of 0.748 or more in the data of the test phase, and even when sparse coding is used, an essentially important weight is maintained. It can be said that. On the other hand, from the viewpoint of calculation load, the decoders A and B have non-zero elements of the weight vector reduced to 36.8% and 15.8%, respectively, compared with the decoder C, and the calculation time can be shortened. The required memory capacity can be reduced.

  In the technology according to Patent Document 1 described in the background art, a wearer-specific dynamic parameter is identified by a parameter identification unit in order to compensate for individual differences of the wearer, and based on an equation of motion in which the identified dynamic parameter is substituted. The drive source was controlled.

  However, in the technique disclosed in Patent Document 1, the wearer-specific dynamic parameters are only used for gravity compensation and inertia compensation, and are not used for the control for estimating the wearer's intention. For this reason, the technique concerning patent document 1 cannot be applied to the control system which estimates a patient's intention continuously and assists like rehabilitation.

  That is, in rehabilitation, it is required to continuously estimate the patient's intention. For this reason, there has been a demand for a motion estimation device and a motion estimation method for reading a patient's intention from a patient's biological signal and estimating the patient's motion in real time based on the biological signal.

  In the technique disclosed in Patent Document 1, a myoelectric potential signal is used as a biological signal. However, the role of the myoelectric potential signal in Patent Document 1 is selection of an operation phase and a standard command output corresponding to the operation phase, and does not continuously and accurately estimate the wearer's movement.

  On the other hand, in the invention according to the present embodiment, a sparse weight vector is generated in advance using the true value of the patient's momentum and the patient's biological signal. And the patient's motion estimated value is produced | generated using the biometric signal detected in real time when the patient is performing rehabilitation, and this sparse weight vector. At this time, in the invention according to the present embodiment, since the sparse weight vector is used when generating the motion estimation value of the patient, the calculation load when calculating the motion estimation value can be reduced. Therefore, even when a biological signal is continuously detected as in rehabilitation, a motion estimation value can be calculated quickly, and a motion estimation value based on the patient's biological signal can be obtained in real time. .

  When sparse coding is used, the number of non-zero elements of the weight vector is reduced. At this time, as described above, even when the sparse coding is used, the weight elements that are essentially important are maintained. Therefore, while reducing the calculation load when obtaining the motion estimation value, the motion estimation value is reduced. Accuracy can be maintained.

  Further, by using the motion estimation device according to the present embodiment, a physical sensor for motion estimation (such as a gyro sensor or an acceleration sensor) or motion capture by image recognition becomes unnecessary, and the motion estimation device can be manufactured at low cost. Can be realized. Moreover, in the motion estimation apparatus according to the present embodiment, the patient's motion is estimated using the upstream biological signal, so that the patient's intention can be quickly estimated.

  In the above-described embodiment, the case where the patient operates the operation unit 11 with the hand (that is, estimation of the upper limb motion) has been described. However, the present invention may be applied to estimation of patient's leg movement. When the present invention is applied to lower limb movement, a myoelectric potential sensor is provided in the muscle of the patient's lower limb. Further, in the above embodiment, the case where the operation unit 11 is displaced on the two-dimensional plane has been described (see FIG. 2). However, in the present invention, the operation unit may be configured to be displaced three-dimensionally.

  When the patient's 19 electroencephalogram (EEG) or cortical electroencephalogram (EcoG) is used as a biological signal, an electroencephalogram signal is input to the preprocessing units 41 and 51 shown in FIG. When using an electroencephalogram (EEG) or a cortical electroencephalogram (EcoG), the preprocessing units 41 and 51 may omit the rectification process.

  Although the present invention has been described with reference to the above embodiment, the present invention is not limited to the configuration of the above embodiment, and those skilled in the art within the scope of the invention of the claims of the present application claims. It goes without saying that various modifications, modifications, and combinations that can be made are included.

DESCRIPTION OF SYMBOLS 1 Motion estimation apparatus 10 Rehabilitation robot 11 Operation part 12 Motor 13 Operation amount detection part 14 Biosignal detection part 15 Control part 16 Motion estimation part 17 Servo amplifier 18 Display part 19 Patient 21 Operation amount 22 Biological signal 31 Cases 32 and 34 Joint 33, 35 Arm 40 Weight vector generation unit 41 Preprocessing unit 42 Time shift operation unit 43 Forward kinematics processing unit 44 Weight vector calculation unit 50 Motion estimation value generation unit 51 Preprocessing unit 52 Time shift operation unit 53 Motion estimation value calculation Part

Claims (6)

An operation unit operated by a patient undergoing rehabilitation;
An operation amount detection unit for detecting an operation amount of the operation unit;
A biological signal detection unit that detects a biological signal of the patient when the patient operates the operation unit;
A movement estimation unit that estimates the movement of the patient using the operation amount detected by the operation amount detection unit and the biological signal detected by the biological signal detection unit;
The motion estimator is
Weight vector generation for generating a sparse weight vector using the true value of the patient's momentum calculated using the operation amount of the operation unit and the patient's biological signal when the patient operates the operation unit And
Using the sparse weight vector generated by the weight vector generation unit and the biological signal detected in real time by the biological signal detection unit when the patient is performing rehabilitation, the motion estimation value of the patient is generated A motion estimation value generation unit that includes:
Motion estimation device.   The motion estimation apparatus according to claim 1, wherein the weight vector generation unit generates the sparse weight vector such that the number of non-zero elements included in the sparse weight vector is equal to or less than a predetermined number. A relational expression y = A _meas θ is established between the true value y of the momentum, the response matrix A _meas generated by performing the time shift operation on the biological signal, and the weight vector θ.
The weight vector generation unit uses sparse coding to determine the weight vector θ using the true value y of the momentum, the response matrix A _meas , and the relational expression, so that the number of non-zero elements is Generate a sparse weight vector less than the weight vector θ,
The motion estimation apparatus according to claim 1. The patient's motion estimation value is y _h , the response matrix generated by performing a time shift operation on the biological signal detected in real time when the patient is performing rehabilitation is A _meas , and the sparse weight If the vector theta _h, the motion estimation value generating unit that generates a motion estimation value of the patient using the relationship y _{h =} a _meas θ h, to any one of claims 1 to 3 The motion estimation apparatus described.   The biological signal detection unit includes at least one of a myoelectric potential sensor that detects a myoelectric potential of the patient's motion site, an electroencephalogram sensor that detects the patient's electroencephalogram, and a cortical electroencephalogram sensor that detects the cortical electroencephalogram of the patient. The motion estimation device according to any one of claims 1 to 4, wherein The operation unit is configured to be displaceable on a two-dimensional plane by the upper limb movement of the patient,
The biological signal detection unit is a myoelectric sensor that detects a myoelectric potential in the patient's greater pectoral muscle, posterior deltoid muscle, biceps long head, and triceps lateral head,
The motion estimation unit estimates an upper limb motion of the patient;
The motion estimation apparatus according to any one of claims 1 to 4.

Images