JPH07223186A - Robot control device - Google Patents

Abstract

PURPOSE:To provide a robot control device detecting the signals corresponding to the activities of a muscle of an operator and capable of controlling a robot with the signals. CONSTITUTION:An electrode 7 is fitted to an arm 1 as an example of s physical portion of an operator, the myoelectric signal is detected by the electrode 7 and amplified by a myoelectric signal amplification section 9, then the pseudo tension is outputted to an equilibrium position estimation section 15 through a time filter 13. The equilibrium position estimation section 15 estimates the articulation angle from the pseudo tension and gives the articulation angle to a controller 17. The controller 17 gives the given articulation angle to a robot 3 as the target orbit to control it.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a robot controller, and more particularly, to a robot or a robot in a virtual reality world in a computer by an myoelectric signal detected according to the activities of a plurality of muscles in a body part of an operator. Virtual body)
The present invention relates to a robot control device capable of controlling the above.

[0002]

2. Description of the Related Art Conventionally, trajectory planning for an articulated industrial manipulator used for a robot is performed online by a human operator using a teaching box on a factory production line. . However, such a trajectory plan is troublesome in that a plurality of trajectories are divided into a number of small portions, and the position of the representative point in the small portions is industrially stored in a manipulator or the like. Therefore, the operator who performs this work requires a skilled technique, and the trajectory planning can be performed only at night when the production line is stopped or on public holidays. Was there.

On the other hand, off-line trajectory planning is
Although theoretically actively researched, this has not been put into practical use due to problems, especially in a situation where a plurality of obstacles having complicated shapes are arranged in a three-dimensional space. The problems are that the cost for image input is very high to perform trajectory planning to perform the desired task, and the situation becomes complicated and the time for orbit calculation explosively increases. To be

On the other hand, in remote control of a robot, remote presence of an operator, and remote work by a communication system and a robot, as means for continuing the motion trajectory and force of an operator, mechanical measurement such as a master arm or a goniometer to be attached to the operator. Devices have been used. However, these mechanical measuring devices become heavier and heavier as the degree of freedom increases, which puts a heavy burden on the operation of the operator. Further, a mechanical measuring device equipped with an actuator for compensating for weight and inertia is not only expensive, but also extremely dangerous because it may harm an operator's body during a runaway and is not generally used.

Therefore, in order to solve the above-mentioned problems, it has been attempted to control the hand of the robot and the artificial hand / leg by using myoelectric signals.

First, in 1948, Norbert Wiener announced the idea of using myoelectric signals, and thereafter, basic research and practical application were advanced. The technology at that stage is
Since the on / off control and the proportional control for controlling the robot with one degree of freedom have been aimed at, the trajectory planning and control of the arm and hand of the robot having multiple degrees of freedom and the artificial arm / prosthesis cannot be performed.

An apparatus for controlling the arms and hands of a robot having multiple degrees of freedom and artificial hands / legs by using myoelectric signals is disclosed in JP-A-51-43888, JP-A-51-63595, and JP-A-5-63595.
As disclosed in JP-A-8-177647 and JP-A-60-212270, many proposals have been made since the announcement of Norbert Winner. However, even if the proposal was made, no specific proposal was made on how to process the electromyographic signal to obtain a signal for controlling the arm and hand of the robot and the artificial hand and leg. Therefore, Ryoji Suzuki et al. (Medical Electronics and Biotechnology, Vol. 7, No. 1, 47-48)
Japanese Patent Laid-Open No. 2-298479 proposes a multi-layer perceptron, which is based on learning identification by a multi-channel myoelectric signal perceptron proposed by pp. 1969) and controlling an artificial hand by an estimated motion. ing.

[0008]

However, in order to recognize the pattern of the myoelectric signal using the multi-layer perceptron and control the arm and hand of the robot and the artificial hand / prosthesis based on the pattern recognition, the following problems occur. Has occurred.
First, robots and artificial hands that need to perform the same actions as humans cannot identify infinite movement patterns. That is, only limited motion patterns are identified. Next, even with a limited motion pattern, the robot does not correctly identify the motion pattern, and a malfunction occurs due to the misrecognition. next,
In order for a robot or a prosthetic hand to accurately recognize a motion pattern, it is necessary to calculate a time average of a myoelectric signal pattern for several tens of milliseconds to several seconds and a power spectrum. Therefore, in this method, even if the robot or the artificial hand accurately executes the motion pattern, the robot or the artificial hand can perform a fast and smooth motion only by a change in the motion of several times per second. I can't.

By the way, the inventor has filed Japanese Patent Application No. 5-1716.
As proposed in No. 88, a robot controller for controlling a robot is proposed by estimating joint torque and motion trajectory in isometric contraction in a horizontal plane from myoelectric signals. However, for example, the degree of freedom of a person's arm is
Although it is said that there are 7 degrees of freedom including shoulders 3 degrees of freedom, elbows 1 degree of freedom, and wrists 3 degrees of freedom, only 2 degrees of freedom in the plane were taken into consideration. Therefore, the movement of the arm itself is constrained within the plane, and the control of the robot accompanying it is also constrained.

Therefore, according to the present invention, a signal corresponding to the activity of the muscle caused by the operator performing a motion with multiple degrees of freedom is detected, and the joint state of the operator is estimated from the signal and given to the robot as a target trajectory. Accordingly, it is to provide a robot controller capable of controlling a robot.

Further, the present invention is to provide a robot controller which takes into consideration the influence of gravity (gravity compensation) in order to control the movement and posture of the robot in a three-dimensional space. .

[0012]

According to a first aspect of the present invention, there is provided a robot controller for controlling a robot on the basis of activities of a plurality of muscles in a body part of an operator, the activities of the plurality of muscles. Detecting means for detecting a signal according to the above, and converting each detection output of the detecting means into a signal representing pseudo tension corresponding to the tension generated by each muscle, and based on the signal, the position state of the joint in the body part of the operator And a control means for controlling the robot based on the position state of the joint estimated by the nonlinear body model.

According to a second aspect of the present invention, the non-linear body model according to the first aspect estimates the position state of the joint in the body part of the operator by expressing it as a joint angle which is a rotation angle from a predetermined reference.

According to a third aspect of the present invention, the control means according to the second aspect determines that the motion of the body part of the operator can be regarded as a continuous change in a balanced equilibrium state, and the joint angle estimated by the non-linear body model. Is given to the robot as a target trajectory and controlled.

According to a fourth aspect of the present invention, the control means of the second aspect responds to the dynamics of the robot in response to the fact that the movement of the body part of the operator cannot be regarded as a continuous change in a balanced equilibrium state. The feature is that the joint angle estimated by the non-linear body model is given as a target trajectory, and the robot is controlled by the target trajectory.

In a fifth aspect, the non-linear body model of the first aspect includes a neural circuit, and the neural circuit filters a signal corresponding to the activities of a plurality of muscles and generates a pseudo tension corresponding to the tension generated by each muscle. It includes a pre-stage circuit for converting into a signal to be expressed, and a post-stage circuit composed of a plurality of layers for estimating the position state of the joint in the body part of the operator from the signal from the pre-stage circuit.

According to a sixth aspect of the present invention, the neural circuit according to the fifth aspect determines a non-linear mapping parameter that is established between a signal corresponding to the activity of a plurality of muscles and the position state of the joint. Learn using an algorithm.

According to a seventh aspect of the present invention, the neural circuit according to the fifth aspect of the present invention uses a predetermined learning rule for determining a parameter of a non-linear mapping that is established between a signal corresponding to the activity of a plurality of muscles and the position state of the joint. Learn using.

In the eighth aspect, the pre-stage circuit of the fifth aspect is 2
It contains a filter formed from two layers.

According to a ninth aspect of the present invention, the latter-stage circuit of the fifth aspect estimates the position state of the joint of the operator's arm, and the position state of the joint of the arm includes a balanced state in which the operator's arm is in a balanced state and has a predetermined value. It is represented by the joint angle, which is the rotation angle from the reference.

According to a tenth aspect of the present invention, the latter-stage circuit of the fifth aspect estimates the position state of the joint of the arm of the operator, and the position state of the joint of the arm includes a state in which the influence of gravity is also taken into consideration and has a predetermined reference. It is represented by the joint angle that is the rotation angle from.

[0022]

According to the first aspect of the present invention, the robot controller detects a signal corresponding to the activity of a plurality of muscles in the body part of the operator and converts it into a signal representing a pseudo tension corresponding to the tension generated by each muscle. The position state of the joint in the body part of the operator can be estimated in real time based on the signal, and the robot can be controlled based on the position state of the joint estimated in that real time.

In the robot controller according to the second aspect of the present invention, the position state of the joint in the body part of the operator is expressed by a joint angle which is a rotation angle from a predetermined reference and estimated, and the joint angle is used as a feature amount. The robot can be controlled based on this.

According to a third aspect of the present invention, there is provided a robot control apparatus, wherein the estimated joint angle is applied to the robot as a target trajectory in accordance with the fact that the motion of the operator's body part can be regarded as a continuous change in a balanced equilibrium state. As a result, high precision control can be performed.

According to another aspect of the robot controller of the present invention, even if the movement of the operator's body part cannot be regarded as a continuous change in a balanced equilibrium state, the dynamics of the robot are adjusted to the estimated joint angle. A target trajectory can be given by giving a coefficient corresponding to, and the target trajectory can be given to the robot for control.

A robot controller according to a fifth aspect of the present invention uses a nerve circuit as a preferred example for estimating the position state of the joint in the body part of the operator,
The robot can be controlled by estimating the position state of the joint in the body part of the operator.

In the robot controller according to the sixth aspect of the present invention, the neural circuit is trained by using a non-linear optimum algorithm to suppress the estimation error of the position state of the joint in the body part of the operator estimated by the neural circuit as much as possible. You can

In the robot controller according to the seventh aspect of the present invention, the neural circuit is made to learn by using a predetermined learning rule, whereby the estimation error of the position condition of the joint in the body part of the operator can be suppressed as much as possible.

According to the eighth aspect of the present invention, the robot controller uses a filter formed of two layers as a pre-stage circuit of the neural circuit, so that the signal representing the pseudo tension corresponding to the tension generated by each muscle of the operator. Can occur.

In the robot controller according to the ninth aspect of the present invention, the latter stage circuit of the nerve circuit can control the robot by representing the position state of the joint in the balanced state where the operator arm is balanced by the joint angle.

In the robot controller according to the tenth aspect of the present invention, the latter stage circuit of the neural circuit can control the robot by estimating the position state of the joint of the operator's arm with the joint angle in consideration of the influence of gravity.

[0032]

First, before describing the embodiments, a model of posture control using an electromyographic signal will be described. It is assumed that the operator fixes the arm, which is one of the body parts, in a certain posture in the three-dimensional space. At this time, if there is no external force, only the force of gravity is applied to the arm. If this relationship is expressed by a vector, the joint angle θ and the motion command u ↑ to each muscle (hereinafter, the vector is indicated by adding ↑) are expressed as a vector, and the relationship shown in the equation (1) is obtained. Here, τ _m ↑ in the equation (1) is a torque generated in a joint by a muscle, τ _g ↑ is a torque applied to each joint by gravity, and g is a gravitational acceleration. If the joint angle θ ↑ which is the solution of the equation (1) when the motion command u ↑ to each muscle is given, the balanced posture is estimated.

The torque τ _m ↑ is determined by the tension and moment arms generated by many muscles of the joint. The moment arm is the distance between the axis of rotation of the joint and the line of action of the muscle. The relationship between the torque τ _m ↑ and the tension and moment arms shows a strong nonlinearity due to the nonlinearity of the relationship between the motion command and the muscle tension, and the nonlinearity depending on the length-tension curve and the joint angle of the moment arm. . Also, h is a function that determines how much torque is applied to each joint due to gravity, and is obtained by a kinematic relationship, and this function is also a non-linear function. From such non-linearity, the joint angle θ, which is the balanced position, is determined as the solution of the equation (1).
↑ is never solved analytically. However, since muscles are also considered to have spring-like properties, there is only one joint angle θ ↑ at which gravity and the potential energy of the spring are minimum, so u ↑ → joint angle θ ↑ is expressed as a map. it is conceivable that.

Further, even when maintaining a constant posture, each muscle tension may change with time, so that the myoelectric signal must be processed with time. Furthermore, even if the muscle tension changes by causing simultaneous activation and changing the myoelectric signal, the joint takes the same posture unless the joint torque changes. Therefore, the mapping in that case has a many-to-one relationship and is a non-linear mapping. Therefore, joint torque is generated from the myoelectric signal of the operator's body part,
A robot controller based on the determination of the joint angle θ ↑, which is the balance position, will be described.

[0035]

[Equation 1]

FIG. 1 is a schematic block diagram showing a robot controller according to an embodiment of the present invention.

Referring to FIG. 1, the robot controller 5 comprises
A surface electrode 7 attached to the human arm 1, a myoelectric signal amplification unit 9 that amplifies an myoelectric signal, a non-linear body model 11 that estimates a joint angle from the amplified myoelectric signal, and a robot 3 is controlled by giving a target trajectory. And a control unit 17 for controlling the operation.

The surface electrode 7 is an example of a detecting means for detecting a signal corresponding to the muscle activity of the operator, for example, a myoelectric signal, and is attached to the human arm 1 which is one of the body parts of the operator. The detected myoelectric signal is amplified by the myoelectric signal amplifier 9 and given to the nonlinear body model 11. The non-linear body model 11 includes a time filter 13 and a balanced position estimation unit 15. Therefore, the time filter 13 filters the amplified myoelectric signal, converts it into a signal representing pseudo tension corresponding to the tension of the human arm 1, and gives it to the equilibrium position estimation unit 15. The equilibrium position estimation unit 15 uses, for example, a non-linear neural circuit, and the human arm 1 based on the signal representing the pseudo tension.
Is estimated and the joint angle when the balance is maintained is given to the control unit 17. The control unit 17 gives the estimated joint angle as a target trajectory to the robot 3 to perform, for example, feedback control.

FIG. 2 is a flow chart for explaining the operation of the robot controller according to one embodiment of the present invention, and FIG. 3 is a flow chart showing the steps of FIG. 2 (5 represented by S in the drawing in more detail). The operation of the robot controller shown in FIG. 1 will be described below with reference to FIGS.

First, steps 1 to 5 will be described. When the surface electrode 7 is attached to the human arm 1 as a body part of the operator and the operator moves the body or generates a force, the myoelectric signal is given to the myoelectric signal amplifying section 9 and amplified. This amplified amplified myoelectric signal is given to the time filter 13, and the signal representing the pseudo tension is given to the equilibrium position estimation unit 15. Next, in steps 51 to 54, the equilibrium position estimation unit 15 obtains the tension from the applied pseudo tension using the length-tension curve. The joint torque is obtained from the obtained tension and the moment arm. A joint angle in a balanced equilibrium state in which the gravity compensation for the human arm 1 is considered is estimated.
The estimated joint angle is given to the control unit 17, and in Step 6, the control unit 17 gives the estimated joint angle as a target trajectory to the robot 3 for control.

Next, the relationship between the myoelectric signal and the pseudo tension required in the operation performed by the time filter 13 will be described. The surface myoelectric signal is a signal in which the membrane excitation potential is temporally and spatially superimposed. The output signal that has passed through the low-pass filter, which is an example of the time filter 13, is expected to reflect the firing frequency of the α motor neuron. This signal is also called pseudo tension because it is considered to be quite close to the true tension. It is known that a low-pass filter is sufficient for a second-order system, and when the input is EMG and the output is T'and an FIR filter is realized, the relationship shown in the equation (2) is obtained. Here, EMG represents a myoelectric signal and T'represents pseudo tension.

Next, the relationship between the motion command and the muscle tension will be described. Between the tension generated by the muscle and the movement command,
It has the following non-linear properties. The tension generated by a muscle increases as the muscle length increases even with the same motion command.
The manner of this change is non-linear and is called the length-tension curve. Further, even with the same motion command, the tension decreases as the muscle length shortening speed increases. The manner of this change is also non-linear, and is called a shortening speed-tension curve. In the robot controller shown in FIG. 1, since the joint angle in the equilibrium state during posture control is estimated, the shortening speed-
The tension curve need not be taken into account.

Next, regarding the relationship between tension and joint torque
explain. Joint torque is the product of tension and moment arm.
Depends on As mentioned above, the moment arm is
It is the distance between the axis of rotation of the joint and the line of action of the muscle. Bend joint
When stretched, muscles can be bent by the skin and bones.
Therefore, this distance is not constant, and the relationship between torque and tension is
Depends on the degree non-linearly. That is, the joint torque is the extensor muscle
Caused by the difference between the torque generated by the
It is determined depending on the ment arm. These relationships
It is formulated as shown in equation (3). Where Τ_j，
θ_j, Α_ij(Θ) is the j-th joint torque and function
Node angle, moment arm of i-th muscle
0) for joints not involved in_iIs the i-th muscle
Muscle tension, Τ _i′ Represents the pseudo tension of the i-th muscle.
The subscript t attached to the upper right means transposition. However, n joints
(1 ≦ j ≦ n) and k muscles (1 ≦ i ≦ k)
It

[0044]

[Equation 2]

FIG. 4 is a diagram for explaining the relationship between the joint torque and the balance position. Hereinafter, the balance position does not change even if the movement command changes if the difference in torque generated by the driving muscle and the antagonist muscle is the same. Explain the flexor muscles.

In FIG. 4, the horizontal axis represents the joint angle and the vertical axis represents the muscle torque. Each of τ _f and τ _e indicates a torque of the extensor muscle and the flexor muscle, and each of u _f and u _e indicates a motion command to the flexor muscle and the extensor muscle. In the flexor, the length of the muscle decreases as the joint angle increases, so the curve is rising to the right. On the other hand, in the extensor muscle, as the joint angle increases, the length of the muscle increases, so that the curve is rising to the right. Further, when the motion command increases so as to change from u _e to u _e ′, these inclinations become large and a larger force is generated.

That is, it is assumed that the arms are balanced at a certain joint angle θ as indicated by the solid line τ _f and the solid line τ _e . At this time, the joint torque for compensating for gravity − τ _g
Occurs as indicated by the solid arrow. From this state, the motion command is changed (u _e → u _e ′, u _f →
It is assumed that the respective tensions of the flexor and extensor muscles are increased by u _f ′) as indicated by the dotted line. Since the joint angle θ, which is the balance position, does not change, the magnitude of the joint torque −τ _g that compensates for gravity does not change from the solid arrow as indicated by the dotted arrow. In this way, if the tensions of the flexor and extensor muscles have the same torque difference, the positions are in the same balance. Even if the torque difference is the same, if the tension generated by each muscle is large, the stiffness is high.

FIG. 5 is a diagram for explaining the degree of freedom from the shoulder of the human arm to the elbow. Since the degree of freedom from the shoulder to the elbow is 4, the links 18, 19, 20, 21 shown in FIG. 5 are provided. The links 18, 19, 20, and 21 represent rotary joints having one degree of freedom, and the rotary axes of the joints are indicated by white arrows on the side of each joint. Further, the starting point where the rotation angle is 0 degree is indicated by a black arrow, and the positive rotation direction is taken from this position in the direction of the right-hand screw.

The shoulder joint is capable of the following movements as a multiaxial joint. The first is a lateral arm elevation (abduction) about the anterior-posterior axis and an exercise (adduction) (θ ₁ ) to attract the elevated arm to the trunk. The second is a motion of turning the upper arm outward and a motion of turning it inward (θ ₂ ) about the vertical axis. Third is to raise the upper arm forward (flexion).
And backward (elevation) (θ ₂ ). Fourth is a motion of twisting the upper arm outward (outer rotation) and a motion of twisting the upper arm around the upper arm (inner rotation) (θ ₃ ). Second and third
Have the same motion mechanically, so that assuming one degree of freedom, the shoulder joint has three degrees of freedom. These three
Of the two degrees of freedom, the link 18 represents the _first degree of freedom by the rotation angle θ ₁ , and the link 19 represents the second degree of freedom by the rotation angle θ _2.
The link 19 represents the third degree of freedom by the rotation angle θ ₃ .

Further, the elbow joint can perform a motion with one degree of freedom, which is a motion (θ ₄ ) for bending and extending the elbow by bending and extending the elbow joint. The link 21 represents this degree of freedom by the rotation angle θ ₄ .

Next, since one shoulder joint has three degrees of freedom,
It is replaced by two links of zero length and three joints with one degree of freedom. Therefore, link the upper arm 1 and link the forearm 2
And At this time, the relationship between the joint torque and the joint angle caused by gravity is expressed as in Expression (4) by rewriting the right side of Expression (1) for each component. Here, each of τ _i and θ _{i in} the equation (4) represents the driving torque and the joint angle of the joint angle, and each of M _i , L _i , and S _i represents the mass, length, and joint of each link. Indicates the length to the center of mass.

[0052]

[Equation 3]

FIG. 6 is a diagram showing a neural circuit model as an example of the nonlinear body model of FIG.

Referring to FIG. 6, the neural circuit forms a four-layer network 22, and includes a first layer 25, a second layer 27, and
It includes a third layer 29 and a fourth layer 31. FI for calculating pseudo tension from myoelectric signals in the first layer 25 to the second layer 27
An R filter (linear conversion circuit) 23 is formed, and a nonlinear conversion circuit 24 is formed from the second layer to the fourth layer 31. First layer 25, second layer 27, third layer 29, fourth layer 31
Each has a neuron. The first layer 25 includes neurons 37, the second layer 27 includes neurons 39, the third layer 29 includes neurons 41, and the fourth layer
Layer 31 includes neurons 43a-43d.

The outputs of the neurons 37 of the first layer 25 are the electromyographic signals EM within a certain time from the past to the present.
Gma (n) ... EMGma (n−N + 1), which is input to each of the neurons 29 of the second layer 27. The number of neurons 39 in the second layer 27 corresponds exactly to the number of nerves, and its output represents the pseudo tension 45. The outputs of the neurons 39 of the second layer 27 are input to the neurons 41 of the third layer 29, respectively. The output of the third layer 29 is the neurons 43a, 43b of the fourth layer 31,
It is input to each of 43c and 43d. The output of the neuron 43a of the fourth layer 31 represents the rotation angle θ ₁ of the link 18 shown in FIG.
Represents the rotation angle theta _2, the output of the neuron 43c represents the rotation angle theta ₃ links 20, the output of the neuron 43d represents the rotation angle theta ₃ links 21.

Formed by the second layer 27 to the fourth layer 31.
In the non-linear conversion circuit 24 shown in FIG.
The non-linear nature of the musculoskeletal system, as it does not include any geometric properties
The question to solve the first (1) which is a nonlinear equation of quality and balance
The subject may be learned according to a predetermined learning rule. Thereby
Neurons in the second layer 27 and neurons in the third layer 29
Of the synaptic connection 33 and the third layer 29 connecting the two
41 and neurons 43a, 43b, 43 of the fourth layer 31
The connection load of the synaptic connection 35 connecting c and 43d (1
Real number) is determined. That is, setting of this coupling load
Includes various neural circuits for a given nonlinear transformation.
A learning algorithm of the road model may be used. Preferred
More specifically, this learning algorithm uses estimated joints.
Angle θ₁, Θ ₂, Θ₃, Θ_FourError back propagation
A seeding learning algorithm may be used.

Each of the neurons 41 of the third layer 29 has a non-linear function based on a sigmoid function, and the third layer 29 is further divided into a plurality of layers depending on the definition of this non-linear function. The conversion circuit 24 is not limited to the three layers formed by the second layer 27 to the fourth layer 31.

Next, the muscle for which the myoelectric signal is measured will be described. The measured myoelectric signals were measured as surface myoelectric signals from the following 12 muscles. Shoulder extensor, flexor,
Anterior deltoid as the abductor, adductor, external rotator, and internal rotator muscles (D
LC), upper deltoid muscle (DLA), posterior deltoid muscle (DL)
S), pectoralis major muscle (PMJ), pectoralis major muscle (TEM), latissimus dorsi (LDO), and biceps long head (BIL) and triceps long head as two joint muscles of the shoulder and elbow ( TRL), and also the extensor of the elbow joint, the brachial muscle (BRC) as flexor, the medial triceps head (TRM), the triceps lateral head (TRA), and the pronation of the pronator (PRT).

The myoelectric signals were obtained by using a pair of silver-silver chloride surface electrodes as the surface electrodes 7 shown in FIG.
Sampled at. The electrodes had a diameter of 10 mm with a distance of 15 mm between the electrodes along the muscle fibers. After full-wave rectification of this signal, the average of every 10 points is calculated, and (EMGa
ve), and as shown in the equation (5), the average of every 5 points was taken and smoothed. This signal is converted to the average myoelectric signal (EMG
ma). Therefore, 200 Hz sampling is performed. If the EMGma of the equation (5) is used as the EMG of the equation (2), the pseudo tension T'is calculated.

[0060]

[Equation 4]

Next, the result of estimating the joint angle from the thus calculated pseudo tension will be described. That is,
The result of estimation of the joint angle by the neural circuit shown in FIG. 6 will be described.

FIG. 7 is a graph showing the relationship between the joint angles θ ₁ , θ ₂ , θ ₃ and θ ₄ estimated by the neural circuit shown in FIG. 6 and the average value of the actually measured joint angles of the operator. In particular, FIG. 7A is a graph for the joint angle θ ₁ , FIG. 7B is a graph for the joint angle θ ₂ , and FIG. 7C is a graph for the joint angle θ ₃ . Yes, FIG. 7D is a graph for the joint angle θ ₄ . Here, the horizontal axis in FIGS. 7A to 7D represents the average value of the measured joint angles in units of radians, and the horizontal axis represents the estimated joint angles in units of radians. .

An index hung from the ceiling was given to the subject, and the subject was instructed to simultaneously activate the muscles for 6 seconds while keeping his / her hand on the index and resting. Then, the surface myoelectric signal is detected by using the silver-silver chloride surface electrode described above, and FIG.
1 for the neuron 27 of the first layer 25 shown in FIG.
The EMGma of the two muscles were entered for 0.5 seconds each. Then, 30 neurons were used for an intermediate layer such as the third layer 39. The non-linear conversion circuit 24 is learned by the back-propagation error method using these neurons.

As a result, the neuron 43a of the fourth layer 31
Estimates θ ₁ shown in FIG.
3b, 43c, and 43d estimated the respective joint angles shown in FIGS. 7B to 7D. Figure 7
The respective results when the simultaneous activation is performed for 5.5 seconds in (a) to (d) of FIG. 7 appear as an average value and a standard deviation. The width of each value along the vertical axis represents the standard deviation, and the central value in the standard deviation is the average value. For example, in FIG. 7A, the average value 49 is located at the center position of the standard deviation 47. What can be said as a result is that if the estimated joint angle and the measured joint angle match, they should be on a straight line,
In particular, the joint angle θ ₂ shown in FIG. 7B is in good agreement.

FIG. 8 is a graph showing the result of controlling a virtual arm having the same dynamics as a human arm, using the result estimated by the neural circuit based on the data shown in FIG. 7 as a target trajectory. Particularly, FIG. 8A is a graph corresponding to the joint angle θ ₁ , FIG. 8B is a graph corresponding to the joint angle θ ₂ , and FIG. 8C is corresponding to the joint angle θ ₃ . 8D is a graph corresponding to θ ₄ .

Referring to FIG. 8, in the control device according to the present invention, the joint trajectory to be estimated is estimated from the myoelectric signal of the operator and the posture of the human arm whose regulation is controlled in the three-dimensional space. Therefore, it can be seen that the target trajectory and the actual trajectory on the start point side and the end point side substantially match. That is,
In the case of controlling the position of the robot continuously on the estimated target trajectory, if it is controlled by a slow motion, it is equivalent to continuously moving the equilibrium state, which is the balanced position. The position estimation is performed quite accurately as shown in. This is because the shortening speed-tension curve is not taken into consideration. Therefore, when the robot is controlled by a motion having some speed between the start point side and the end point side, the target trajectory and the actual trajectory vary in the intermediate process as shown in FIG.

However, even if the joint angle, which is the estimated position, fluctuates to some extent, by considering the difference between the robot dynamics and the operator dynamics,
A smooth trajectory can be realized. In addition to considering the difference in dynamics, if the control unit 17 shown in FIG. 1 controls the robot 3 by control such as feedback control or feedforward control, the equilibrium state, which is the balance position, can be continuously changed. The robot can be controlled even when the motion is not changed to 1. By combining these controls, the robot controller according to the present invention can control the robot more naturally.

In this embodiment, the human arm is focused on as the body part of the operator, but the robot is not limited to the human arm, and the joint angle of the body part such as a leg is estimated to control the robot. Good.

In this embodiment, the surface electrode is used as an example of the detection signal for detecting the myoelectric signal, but a needle electrode may be used as another detection means, and only a pair of silver and silver chloride may be used as the electrode. Surface electrodes are considered preferred. here,
What should be taken into consideration is that the detection means does not harm the human body.

Further, the idea used in the robot control device according to the present invention is not limited to the robot, and is considered to be applied to, for example, artificial hands / legs.

Further, the learning rule for the learning performed by the neural circuit is not limited to the error backpropagation learning, and may be the EM algorithm or the optimal algorithm such as the union reward penalty learning.

Further, the control method for the control unit to control the robot is not limited to the feedback control or the feedforward control.

[0073]

As described above, according to the present invention, a signal corresponding to the activity of a plurality of muscles in the body part of the operator is detected, and the detected signal represents the pseudo tension corresponding to the tension generated by each muscle. It is desired to control the robot by continuous motion in balanced equilibrium state, since it is converted into a signal and the position state of the joint of the operator is estimated based on the signal by the non-linear body model to control the robot. It can be controlled by the orbit.

[Brief description of drawings]

FIG. 1 is a schematic block diagram showing a robot controller according to an embodiment of the present invention.

FIG. 2 is a flow chart for explaining the operation of the robot controller according to the embodiment of the present invention.

FIG. 3 is a flowchart showing in more detail a step (denoted by S in the drawing) 5 of FIG.

FIG. 4 is a diagram for explaining a relationship between joint torque and a balance position.

FIG. 5 is a diagram for explaining a degree of freedom from a shoulder to an elbow of a human arm.

FIG. 6 is a diagram showing a neural circuit model as an example of the nonlinear body model of FIG.

7 is a graph showing a relationship between joint angles θ ₁ , θ ₂ , θ ₃ , θ ₄ estimated by the neural circuit model shown in FIG. 6 and actually measured joint angles of an operator.

8 is a graph showing a result of controlling a virtual arm having the same dynamics as a human arm with a joint angle estimated by the neural circuit model shown in FIG. 6 as a target trajectory.

[Explanation of symbols]

 3 Robot 5 Robot controller 7 Surface electrode 11 Non-linear body model 13 Time filter 15 Equilibrium position estimation part 17 Control part 22 Network 23 FIR filter (linear conversion circuit) 24 Non-linear conversion circuit 45 Pseudo tension

Claims (10)

[Claims] 1. A robot control device for controlling a robot based on the activities of a plurality of muscles in an operator's body part, wherein the detection means detects a signal according to the activities of the plurality of muscles, and A non-linear body model that converts each detection output into a signal representing pseudo tension corresponding to the tension generated by each muscle, and estimates the position state of the joint in the body part of the operator based on the signal, and the non-linear body model A robot control device comprising: a control unit that controls the robot based on the estimated position state of the joint. 2. The nonlinear body model according to claim 1, wherein a position state of a joint in a body part of the operator is expressed and estimated by a joint angle which is a rotation angle from a predetermined reference. Robot controller. 3. The robot controls the joint angle estimated by the nonlinear body model in response to the movement of the body part of the operator being regarded as a continuous change in a balanced equilibrium state. The robot control device according to claim 2, wherein the robot control device is controlled by giving it as a target trajectory. 4. The control means gives a coefficient corresponding to the dynamics of the robot when the movement of the body part of the operator cannot be regarded as a continuous change in a balanced equilibrium state. The robot controller according to claim 2, wherein the joint angle estimated by the nonlinear body model is set as a target trajectory, and the robot is controlled by the target trajectory. 5. The non-linear body model includes a neural circuit, and the neural circuit filters a signal corresponding to the activities of the plurality of muscles and converts the signal into a signal representing a pseudo tension corresponding to a tension generated by each muscle. The robot control device according to claim 1, further comprising: a pre-stage circuit for performing the pre-stage circuit, and a post-stage circuit including a plurality of layers for estimating a position state of a joint in the body part of the operator from a signal from the pre-stage circuit. 6. The neural circuit uses a predetermined non-linear optimum algorithm to determine a parameter of a non-linear mapping that is established between a signal according to the activities of the plurality of muscles and a position state of the joint. Characterized by learning,
The robot controller according to claim 5. 7. The neural circuit learns using a predetermined learning rule in order to determine a parameter of a non-linear mapping that is established between a signal according to the activities of the plurality of muscles and a position state of the joint. The robot controller according to claim 5, wherein: 8. The robot controller according to claim 5, wherein the pre-stage circuit includes a filter formed of two layers. 9. The post-stage circuit estimates a position state of a joint of the operator's arm, and the position state of the joint of the arm includes a balanced state of balance of the operator's arm, and from a predetermined reference. The robot control device according to claim 5, wherein the robot control device is represented by a joint angle that is a rotation angle of. 10. The post-stage circuit estimates the position state of the joint of the operator's arm, and the position state of the joint of the arm includes a state in which the influence of gravity is also taken into consideration, and a rotation angle from a predetermined reference. The robot control device according to claim 5, wherein the robot control device is represented by a joint angle.