JPH06114768A - Robot control device - Google Patents

Abstract

PURPOSE:To provide a robot control device which is designed to cope with an error factor, being heretofore non-considerable, improve correction precision, and obviate the need for teaching at a site. CONSTITUTION:An objective position computed by an Off-line teaching host system 60 is received by the objective position Mm' holding part 51 of a robot control device 50 and the objective position Mm' is corrected by a correcting part 52 by means of the output of a neutral network. A robot 10 is controlled by a robot control part 54 based on the objective position Mm' corrected by means of the output of the neutral network.

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 controller in an offline teaching system that corrects the position of a robot by using a neural network.

[0002]

2. Description of the Related Art An absolute positioning error in an industrial robot is caused by a machining error, an assembly error, a dimensional error of a robot arm due to thermal strain, and a flexure of each joint due to its own weight. In the control of the industrial robot in the conventional offline teaching system, as shown in FIG. 11, first, a target position is generated on the side of the offline teaching system 700 (701), and this target position Mm is held (702), Then, this value is corrected by a mathematical model considering the arm length, the assembly angle error, etc. (70
3) The corrected target position Mm 'is calculated, and this is held by, for example, writing it on the floppy disk (7).
04). Then, when the operator operates the robot 710, the floppy disk is moved to the robot controller 7
Move to 50 side. The robot control device 750 reads the corrected target position Mm ′ written in the floppy disk (751) and controls the robot 710 (754).

[0003]

However, even if the target position Mm is corrected on the side of the offline teaching system by a mathematical model in consideration of the error factors, there are various causes of the positioning error of the robot, and the mathematical model is used. There is a limit to the correction accuracy because it is difficult to completely express with.

Further, since the absolute positioning accuracy of the robot is low as described above, when the robot controller of the offline teaching system is set, it is necessary to correct the teaching at a place where it is actually used. For this reason, it takes time to set the robot control device, and a great deal of labor is required to do this, which causes a cost increase.

The present invention has been made to solve the above problems, and an object of the present invention is to provide a robot controller having high correction accuracy and requiring no on-site teaching correction.

[0006]

The structure of the invention for solving the above-mentioned problems is a robot controller for an off-line teaching system, and a target position holding means for receiving a robot control target position calculated by the off-line teaching system. A neural network that outputs a correction amount using the target position as an input; a correction unit that corrects the calculated target position by the output of the neural network; and a robot control based on the target position corrected by the neural network. And a control means for performing.

[0007]

According to the above means, the target position holding means receives the target position calculated by the offline teaching system, the correction means corrects the calculated target position by the neural network, and the control means operates by the neural network. The robot is controlled based on the corrected target position.

[0008]

EXAMPLE A robot controller according to this example will be described below with reference to the drawings. First, with reference to FIG. 3, an outline of the operation of the robot control device 50 according to the present embodiment and the offline teaching host system 60 that gives the robot control target position to the robot control target position will be described. On the side of the offline teaching system 60, the target position generation unit 61 generates the position matrix Mm of the robot control target, the target position Mm holding unit 62 holds this target position matrix Mm, and the correction unit 63 uses this target position matrix. The value of Mm is the arm length of the robot,
A corrected target position matrix Mm ′ is obtained by performing correction using a mathematical model that takes into account the assembly angle error and the like. next,
The target position holding unit 64 writes this as control information of the robot, for example, on a floppy disk. Then, the operator takes out the floppy disk in which the corrected target position matrix Mm ′ is written from the target position matrix Mm ′ holding unit 64 and sets it in the target position matrix Mm ′ holding unit 51 on the robot controller 50 side. Thus, the robot control device 50 is operated. The target position holding unit 51 of the robot controller 50 is
The corrected target position matrix Mm ′ written on the floppy disk is read out, and the correction unit 52 outputs the target position matrix M based on the output of the neural network.
The target position matrix Mm ″ that is further corrected by the neural network is obtained by correcting m ′. The target position matrix Mm ″ holding unit 53 holds this, and the robot control unit 54 controls the robot 10 based on the target position matrix Mm ″ corrected by this neural network. The robot 10 having 6 joints controlled by the robot controller 50 will be described in more detail with reference to FIG. The robot 10 has a column 1 attached to a pedestal 12 fixed to a base 13.
4, a first arm 15, a second arm 16, a third arm 17, and a finger 19, and the first joint a,
The second joint b, the third joint c, the fourth joint d, the fifth joint e, and the sixth joint f are configured so that the position and posture of the finger 19 can be freely controlled with 6 degrees of freedom.

Next, the configuration of the robot controller 50 for controlling the robot 10 will be described with reference to FIG. An operating box 27 for inputting a control command is input to the CPU 11 that calculates and controls the position of the robot 10.
An operation panel 26, an external storage device 29 for holding control information (floppy disk) created by the offline teaching host system 60 shown in FIG. 3, a ROM 20 described later in detail, and a RAM 30 are connected. Further, the CPU 11 is connected to the 1-axis servo control section 40a for controlling the first joint a through the 6-axis servo control section 40f for controlling the sixth joint, and the CPU 11 is housed in the external storage device 29 described above. The processing set based on the control information is performed, and a control command is issued to the 1-axis servo control section 40a to 6-axis servo control section 40f. In response to this, each of the servo control units (40a to 40f) drives the robot 10 by rotating the servo motors M1 to M6 and moving the first joint a to the sixth joint f. Each servo motor M1 to M
The movements of 6 are fed back to the respective servo control units by the encoders E1 to E6.

Next, the position correction of this embodiment will be described prior to the description of the control by the robot controller. In the control of the conventional industrial robot, as described above in the section of the related art with reference to FIG. 11, the target position matrix Mm of the tip of the robot arm is obtained (701) on the offline teaching system 700 side, and The position error correction is performed on the target position matrix Mm using a mathematical model d in consideration of the error, the corrected target position matrix Mm ′ is obtained (703), and this is held (704). Then, on the robot controller 750 side, the corrected target position matrix Mm ′ is inversely transformed by the inverse transformation function fa to obtain the joint angle vector Θa which is the control target of each joint of the robot, and the robot is based on this. Were controlled (754).
The position matrix Ma of the robot arm tip, which is controlled based on the joint angle vector Θa, is obtained by forward transforming the joint angle vector Θa by an actual forward transformation function gt that reflects all errors.

## EQU00001 ## d fa gt Mm.fwdarw.Mm'.fwdarw..THETA.a.fwdarw.Ma Equation 1 where Mm: target position matrix d: mathematical model in which error is limited Mm ': target position corrected by mathematical model d Matrix fa: Inverse transformation function Θa: Joint angle vector to be controlled gt: Actual forward transformation function that reflects all errors Ma: Position matrix of robot arm tip controlled based on Θa

Here, the robot is controlled by obtaining the joint angle vector Θa on the basis of the target position matrix Mm ′ corrected by the mathematical model d in which the error factors are limited and taken into consideration. Position matrix Ma
Contains an error that could not be considered in the mathematical model d, and this error became a control error of the robot controller. That is, the forward conversion function gt that defines the actual position matrix Ma contains an error that could not be considered, and the difference caused by the correction by the mathematical model d and the forward conversion by the forward conversion function gt becomes an error, which is controlled. Further, there is a position error between the position matrix Ma of the robot arm and the target position matrix Mm.

On the other hand, in the present embodiment, the configuration as shown in FIG. 3 is adopted so that the position matrix Ma of the controlled robot arm shown in Expression 1 becomes equal to the target position matrix Mm. The target position matrix Mm ′ corrected by the mathematical model d is further corrected by a correction matrix ΔMm (output of the neural network) which is an error amount (necessary correction amount) (52), and the corrected target position matrix Mm ′ is obtained. The robot 10 is controlled on the basis of '(54). The concept of this correction will be described in more detail below.

First, it is considered that the error amount caused by the correction by the mathematical model d and the forward conversion by the forward conversion function gt corresponds to the necessary correction amount, and the correction by the mathematical model d and the forward conversion by the forward conversion function gt are performed as a correction matrix. When ΔMm is set, Expression 1 can be expressed by the following expression.

## EQU00002 ## Mm..DELTA.Mm = Ma Equation 2 Then, from Equation 2, the error amount (correction matrix) .DELTA.Mm is expressed by the following equation.

## EQU3 ## ΔMm = Mm ⁻¹ · Ma Equation 3

Here, the error amount (correction matrix) ΔMm
The inverse matrix ΔMm ^{−1 of} is regarded as a necessary correction amount, and in the present embodiment, the target position matrix Mm ′ corrected by the offline teaching host system is multiplied by this correction amount ΔMm ⁻¹ (Mm ′ · ΔMm ⁻¹ ). Is a target position matrix Mm ″ corrected by the neural network, and the robot is controlled based on this.

Next, the robot controller 5 according to this embodiment.
The learning of 0 into the neural network will be described with reference to FIG. 4, which is a block diagram showing the outline of the arithmetic processing of the CPU 11. First, the target position matrix Mm ′ input unit 401
Then, the position matrix Mm 'of the control target calculated by the offline teaching host system 60 (this is the target position matrix Mm first generated as shown in FIG. 3 and corrected by using the mathematical model d considering the error). ) Is entered. Then, the coordinate inverse transformation fa computing unit 403 computes the joint angle vector Θa of each joint of the robot corresponding to the target position matrix Mm ′ by the inverse transformation function fa.

Next, the joint angle vector Θa of each joint of the robot obtained by the coordinate inverse transformation fa computing unit 403 is held in the joint angle vector Θa holding unit 405, which becomes an error factor of robot control at this time. The load, the temperature, etc. are held in the error factor item holding unit 407. Then, based on the joint angle vector Θa, the robot controller 411 controls the robot 1
0 is operated. After this operation, the position matrix Ma measuring unit 413 measures the actual tip position matrix Ma of the robot arm. Then, the correction matrix ΔMm calculation unit 415 obtains an inverse matrix of the target position matrix Mm ′, and using the above-mentioned Equation 3 (ΔMm = Mm ⁻¹ · Ma) from the inverse matrix and the tip position matrix Ma, The correction matrix ΔMm is obtained. Then, the calculated correction matrix ΔMm is stored in the correction matrix ΔMm holding unit 417.
To hold.

Accumulation of correction data by the above processing is performed for a plurality of positions. Then, the neural network 409 is made to perform learning using the accumulated joint angle vector Θa and the error factor item as input data and the correction matrix ΔMm as teacher data, and the neural network 409 has the relation of the correction matrix ΔMm with respect to the joint angle vector Θa. To learn. The learning of the neural network 409 will be described in more detail later.

Next, the robot controller 5 according to the present embodiment.
0, neural network 4 that has completed the above learning
The control using 09 will be described with reference to the block diagram of FIG. 5 showing the outline of the calculation by the CPU 11. First, the target position matrix Mm ′ input unit 501 acquires the control target position matrix Mm ′ of the robot 10 calculated by the offline teaching host system 60 (values corrected by the mathematical model d). Then, the first coordinate inverse transformation fa calculation unit 503 obtains the joint angle vector Θa of each joint of the robot 10 using the inverse transformation function fa.

Next, the joint angle vector Θa of each joint of the robot 10 obtained by the first inverse coordinate transformation fa calculation unit 503.
Is held in the joint angle vector Θa holding unit 505, and at the same time, the error factor item holding unit 507 holds the load, temperature, and the like, which are error factors in robot control at this time. And
The joint angle vector Θa and the error factor item are input to the neural network 409, and the joint angle vector Θa is input.
Also, the correction matrix ΔMm for the error factor at this time is acquired and held in the correction matrix ΔMm holding unit 517. Then, the target position correction unit 519 obtains an inverse matrix ΔMm ⁻¹ of the correction matrix ΔMm held in the correction matrix ΔMm holding unit 517, and this is used as the target position matrix Mm ′ held in the target position matrix input unit 501. By multiplying, the corrected target position matrix Mm ″ (Mm ″ = Mm ′ · ΔMm ⁻¹ ) is obtained. Finally, in the second coordinate inverse transformation fa calculation unit 521, based on the corrected target position matrix Mm ″, the inverse transformation function fa is used to joint angle vectors of the first joint a to the sixth joint f of the robot 10. Find Θa '. Then, the position control unit 523
On the basis of the joint angle vector Θa 'thus obtained, issues a command to the 1-axis servo control units 40a to 6f shown in FIG. 2 described above, and in response thereto, the 1-axis servo control units 40a to 6a. The axis servo control unit 40f is the servo motor M1.
By driving M6 to M6, the first to sixth joints a to f are moved, and the position and posture of the robot 10 are controlled.

In this embodiment, the neural network 4
When learning in 09, the joint angle vector Θa is obtained from the target position matrix Mm ′, this joint angle vector Θa is given to the neural network 409, and the correction amount for the target position matrix Mm ′ on which the joint angle vector Θa is calculated ( Since the correction matrix ΔMm) is obtained, there is an advantage that learning is simple compared with the configuration in which the correction matrix is obtained directly from the target position matrix. Further, in this embodiment, since the joint angle vector Θa is given to the neural network to obtain the correction matrix ΔMm, the error factor can be reduced.
This is because there are a plurality of joint angle vectors Θa with respect to the target position matrix Mm ′, and therefore, when the target position matrix is directly given to the neural network, the plurality of joint angle vectors Θa become an error factor. This is because the joint angle vector Θa is given to the neural network, and the robot has only one tip position determined by the joint angle vector Θa.

Next, the learning of the neural network 409 of the robot controller according to the above-described embodiment and the calculation by the learning will be described in more detail. 1. Configuration of Neural Network, The neural network of the present embodiment has the CP shown in FIG.
The computer system is composed of U11, ROM 20, and RAM 30. The ROM 20 stores a control program area 21 in which a control program for managing the accumulation of input data and teacher data is stored, a neural network area 22 in which a neural network operation program is stored, and a program for learning the neural network A learning program area 23 is formed. In the RAM 30, the joint angle vector Θa stored in the joint angle vector Θa holding unit 405 described above with reference to FIG. 4 and the error factor item holding unit 407.
An input data storage area 31 for storing error items such as temperature and load accumulated as input data, and a teacher data storage area 32 for similarly storing the correction matrix ΔMm stored in the correction matrix ΔMm holding unit 417 as teacher data. And a coupling coefficient storage area 33 for storing the coupling coefficient of the neural network.

2. Neural Network As shown in FIG. 7, the neural network 409 of this embodiment has a three-layer structure of an input layer LI, an output layer L0, and an intermediate layer LM. The input layer LI has e input elements, the output layer L0 has g output elements, and the intermediate layer LM
Has h output elements.

A multi-layered neural network is generally defined as a device that performs the following operations. The output O ⁱ _j of the j-th element in the i-th layer is calculated by the following equation.
However, i ≧ 2.

Equation 4 O ⁱ _j = f (I ⁱ _j ) Equation 4

[Equation 5]

## EQU00006 ## f (x) = 1 / {1 + exp (-x)} Expression 6

However, V ⁱ _j is the bias of the j-th arithmetic element of the ^i- th layer, and W ^i-1 _k, ⁱ _j is the k-th element of the i-1-th layer and the j-th element of the i-th layer. Coupling coefficient of each of the th element, O ¹ _j
Represents the output value of the j-th element in the first layer. That is, the first
Since it is a layer, the input is output as it is without performing any calculation, so it is also the input value of the j-th element of the input layer (first layer).

Next, a specific calculation procedure of the neural network 409 having the three-layer structure shown in FIG. 7 will be described with reference to FIG. The calculation of each layer is performed by referring to the coupling coefficient stored in the coupling coefficient storage area 33 of the RAM 30 and executing the program stored in the neural network area 22 of the ROM 20.
In step 100, the j-th element of the intermediate layer (second layer) is the output value O ¹ from each element of the input layer (first layer).
_j (input data of the first layer) is input, and the product-sum calculation of the following equation, which is an embodiment of Equation 5 using the layer number and the number of elements of the first layer, is performed.

[Equation 7]

Next, at step 102, the output of each element of the intermediate layer (second layer) is calculated by the following equation using the Sigmond function of the product-sum function value of the input values of equation 7. The output value of the j-th element in the second layer is calculated by the following equation.

Equation 8] _{^{O 2 j = f (I 2}} j) = 1 / {1 + exp (-I 2 j)} Equation 8 The output value O ² _j is the input value of each element of the output layer (third layer) .

Next, at step 104, the sum of products operation of the input values of the respective elements of the output layer (third layer) is executed.

[Equation 9] Next, in step 106, the output value O ³ _j of each element of the output layer is calculated by the Sigmont function as in the case of the equation 8.

## EQU10 ## O ³ _j = f (I ³ _j ) = 1 / {1 + exp (-I ³ _j )} Equation 10

3. Structure of input data and teacher data The data used for the update learning of the neural network is configured in the database as shown in FIG. Input data is D ₁ ··· D _n, the corresponding teacher data is E ₁ ··· E _n. The n pieces of input data are the joint angle vector Θa held in the joint angle vector Θa holding unit 405 described above with reference to FIG. 4 and the error factor temperature held in the error factor item holding unit 407,
The weight and the like, and the n pieces of teacher data are the correction matrix ΔMm held in the correction matrix ΔMm holding unit 417. And each of these data
It is stored in the input data storage area 31 and the teacher data storage area 32 of the RAM 30.

This input data is defined as follows.
Consider the e pieces of data given to each of the e input elements as one set of data. Then, an arbitrary m-th set of input data is represented by D _m , and input data for the j-th input element belonging to the set is represented by d _mj . D _m represents a vector, and d _mj is a component of the vector. That is, D _m is defined by the following equation.

[Equation 11] D _m = (d _m1 , d _m2 , ... D _me-1 , d _me ) Equation 11 Further, n sets of input data are D ₁ , D ₂ , ... D _n-1 , D
It is represented by _n . Hereinafter, all n sets of input data groups are referred to as an input data group D. When the expression 7 is used for the input data D _m , the component d _mk is substituted into O ¹ _{k of the} expression 7.

Similarly, E ₁ , ... E _n are defined as follows. For the output layer LO, consider the teacher data for the output from each of the g output elements as a set of data. Then, an arbitrary m-th set of teacher data is represented by E _m , and teacher data for the j-th output element belonging to the set is represented by _em j. E _m represents a vector and e _mj is a component of the vector. That is, _Em is defined by the following equation.

[Equation 12] E _m = (e _m1 , e _m2 , ... E _mg-1 , e _mg ) Equation 12 Further, n sets of teacher data are E ₁ , E ₂ ,.
It is represented by E _n-1 and E _n . Hereinafter, all n sets of teacher data groups will be referred to as a teacher data group E.

4. Learning Neural Network This neural network performs RO as initial learning.
Learning is performed by executing the program of the procedure shown in FIG. 9 stored in the learning program area 23 of M20. The learning of the coupling coefficient is performed by the well-known backpropagation method.

This learning is repeatedly executed with respect to a large number of input data regarding various events so that the respective outputs become the respective optimum teacher data. These input data and teacher data are stored in the input data storage area 31 and the teacher data storage area 32, respectively, as described above.

In step 200 of FIG. 9, the data number i is set to the initial value 1, and the output element number j (the teacher data component number j) is set to the initial value 1. Next, in step 202, the i-th input data Di and the i-th teacher data Ei are extracted from the input data storage area 31 and the teacher data storage area 32.

Next, in step 206, the learning signal of the element corresponding to the j-th component e _ij of the i-th teacher data Ei read out from the output layer is calculated by the following equation.

[Formula 13] Y ³ _j = (e _ij −O ³ _j ) · f ′ (I ³ _j ) Formula 13 However, in Y ³ _j , O ³ _j , and I ³ _j , the data number i is omitted. f '(X) is the derivative of the Sigmond function. Further, I ³ _j is the O of the equation 7 in which each component of the input data D _i is
^By substituting ¹ _k, I ² _k is obtained for all the elements in the intermediate layer, I ² _k is substituted in the equation 8 to obtain outputs O ² _k for all the elements in the intermediate layer, and O ² _k is obtained for all the k. ^It is calculated by substituting ² _k into Equation 9. In addition, O ³ _j is obtained by using I ³ _j in Equation 1
It is obtained by substituting 0.

Next, at step 210, it is judged whether or not learning signals have been calculated for all output elements,
If the determination result is NO, in step 212,
The element number j is incremented by 1, and the process returns to step 206,
A learning signal for the next output element is calculated. When it is determined in step 210 that the calculation of the learning signal for all the output elements is completed, in step 214, the learning signal Y for an arbitrary r-th element in the intermediate layer is calculated by the following equation.

[Equation 14] Such a learning operation is performed on all the elements in the middle layer.
Be done.

Next, in step 216, each coupling coefficient of the output layer is corrected. The correction amount is calculated by the following equation.

[Expression 15] Δω ² _i, ³ _j (t) = P · Y ³ _j · f (I ² _i ) + Q · Δω ² _i, ³ _j (t−1) Equation 15 where Δω ² _i, ³ _j (T) is the amount of change in the t-th calculation of the coupling coefficient between the j-th element of the output layer and the i-th element of the intermediate layer. Also, Δω ² _i, ³ _j (t-1) is
It is the previous correction amount of the coupling coefficient. P and Q are proportional constants. Therefore, the coupling coefficient is

## EQU16 ## W ² _i, ³ _j + Δω ² _i, ³ _j (t) → W ² _i, ³ _{j The} corrected coupling coefficient is obtained by the equation 16.

Next, in step 218, each coupling coefficient of each element of the intermediate layer is corrected. The correction amount of the coupling coefficient is obtained by the following equation, as in the case of the output layer.

Δω ¹ _i, ² _j (t) = P · Y ² _j · f (I ¹ _i ) + Q · Δω ¹ _i, ² _j (t−1) Equation 17 Therefore, the coupling coefficient is

[Expression 18] W ¹ _i, ² _j + Δω ¹ _i, ² _j (t) → W ¹ _i, ² _{j The} corrected coupling coefficient is obtained by the expression 18.

Next, at step 220, it is judged whether or not one learning is completed for the n pieces of input data and the teacher data to be learned. If learning for all input data has not been completed, the process proceeds to step 222 to read the next input data and the teacher data corresponding to the input data from the input data storage area 31 and the teacher data storage area 32. The data number i is incremented by 1, and the component number j is set to the initial value 1. And
Returning to step 202, the above learning is executed using the next input data and the teacher data.

When it is determined in step 220 that learning has been completed for all n input data and teacher data,
In step 224, it is determined whether the coupling coefficient has converged by determining whether the square value of the difference between the output data and the teacher data has become equal to or less than a predetermined value. If the coupling coefficient has not converged, the process returns to step 200, and the learning described above is executed from the first input data and the teacher data in order to perform the second learning.

In this way, in step 224, the above learning operation is repeatedly executed until the square value of the difference between the output data and the teacher data becomes equal to or less than the predetermined value and the learning converges. This completes the initially learned neural network for a wide range of early events. As a result of this learning, the neural network 409 of the present embodiment can calculate the necessary correction matrix ΔMm by inputting the joint angle vector Θa as shown in FIG.

Next, the learning of the neural network of the robot controller according to another embodiment of the present invention will be described with reference to FIG. 6 which is a block diagram showing the outline of the arithmetic processing of the CPU. In the above-described embodiment, the joint angle vector Θa is input to the neural network 409 as input data.
Then, the correction matrix ΔMm is learned as the teacher data, but in the present embodiment, the neural network 409 is made to learn the target position matrix Mm ′ as the input data and the correction matrix ΔMm as the teacher data. In this embodiment as well, the configuration of the robot controller is substantially the same as that of the embodiment described above with reference to FIGS. 2 and 3, and therefore its explanation is omitted.

First, the target position matrix Mm 'input unit 6
At 01, the position matrix Mm ′ of the control target of the robot calculated by the offline teaching host system 60 using the mathematical model d considering the error is input. Then, the coordinate inverse transformation fa computing unit 603 computes the joint angle vector Θa of each joint of the robot corresponding to the target position matrix Mm by the inverse transformation function fa. Next, the target position matrix Mm of the target position matrix Mm ′ input unit 601
The target position matrix Mm ′ holding unit 605 holds m ′, and the error factor item holding unit 607 holds the load, the temperature, and the like, which are error factors of the robot control at this time. Then, the robot controller 611 operates the robot 10 based on the joint angle vector Θa described above. After this operation, the position matrix Ma measuring unit 613 measures the actual tip position matrix Ma of the robot arm. And
The correction matrix ΔMm calculation unit 615 obtains the inverse matrix of the target position matrix Mm, and from the inverse matrix and the tip position matrix Ma, the above-mentioned expression 3 (ΔMm = Mm
⁻¹ · Ma) is used to find the correction matrix ΔMm. Then, the obtained correction matrix ΔMm is held in the correction matrix ΔMm holding unit 617.

Accumulation of correction data by the above processing is performed for a plurality of positions. Then, the neural network 409 is caused to perform learning using the accumulated target position matrix Mm ′ and error factor items as input data and the correction matrix ΔMm as teacher data, and the neural network 4
09 is made to learn the relation of the correction matrix ΔMm to the target position matrix Mm ′. The control of the robot that performs the correction using the learned neural network 409 is substantially the same as the embodiment described above with reference to FIG. As described in the embodiment shown in FIG. 6 and the embodiment described above with reference to FIG.
In the present invention, various data can be used as the learning data of the neural network 409.

In this embodiment, a neural network having a three-layer structure consisting of the input layer LI, the intermediate layer LM, and the output layer LO is taken as an example, but the neural network of the present invention is not limited to such a configuration and is necessary. The desired purpose of the present invention can be achieved by a neural network having any configuration capable of performing various learning. Further, when the concept of the correction of the present embodiment is described, mathematical expressions are given for convenience of explanation, but the present invention is not limited to these mathematical expressions.
Further, in the above description with reference to FIG. 3, the control data of the robot calculated by the offline teaching host system 60 was transferred to the robot controller 50 via the floppy disk, but the data transfer is performed via the data line or the like. It is also possible.

[0045]

The present invention is configured as described above and corrects the error factors, which cannot be taken into consideration by the conventional mathematical model, by learning them by the neural network, and therefore the correction accuracy can be improved. is there. Also,
Since the robot correction accuracy is high, it is not necessary to make teaching corrections at the actual location where the robot controller is set, so it is possible to shorten the robot controller setting time. The cost of setting the control device can be reduced. Further, there is an advantage that the safety can be improved.

[Brief description of drawings]

FIG. 1 is a configuration diagram showing a mechanical configuration of a robot according to an embodiment of the present invention.

FIG. 2 is a block diagram showing the configuration of a robot control apparatus according to this embodiment for controlling the robot shown in FIG.

FIG. 3 is a block diagram showing the configurations of an offline teaching host system and a robot controller according to the present embodiment.

FIG. 4 is a block diagram illustrating processing during learning of the robot control apparatus according to the present embodiment.

FIG. 5 is a block diagram illustrating a control process after learning of the robot controller according to the present embodiment.

FIG. 6 is a block diagram illustrating a control process during learning of a robot controller according to another embodiment of the present invention.

FIG. 7 is a configuration diagram showing a configuration of a neural network of the robot controller according to the present embodiment.

8 is a flowchart showing a calculation procedure of the neural network according to the embodiment shown in FIG.

9 is a flowchart showing a learning procedure of the neural network according to the embodiment shown in FIG.

FIG. 10 is a configuration diagram showing a data configuration of a database having input data and teacher data used for learning of a neural network.

FIG. 11 is a block diagram showing the configurations of a conventional offline teaching host system and a robot controller.

[Explanation of symbols]

10 Robot 11 CPU 20 ROM 30 RAM 50 Robot Controller 51 Corrected Target Position Matrix Mm ′ Holding Unit 52 Neural Network Correcting Unit 53 Neural Network Corrected Target Position Mm ″ Holding Unit 54 Robot Control Unit 60 Offline Teaching System 61 Target position generating unit 63 Correcting unit using mathematical model 409 Neural network LI input layer LM intermediate layer LO output layer

Claims (1)

[Claims] 1. A robot controller for an offline teaching system, comprising: target position holding means for receiving a target position for robot control calculated by the offline teaching system; and a neural network for outputting a correction amount with the target position as an input. A robot controller, comprising: a correction unit that corrects the calculated target position based on the output of the neural network; and a control unit that controls the robot based on the target position corrected by the neural network.