JPS61253505A - Learning controller for robot arm - Google Patents

Abstract

PURPOSE:To reduce less an error with each repetition of trials and to improve the performance of the learning control of a robot arm, by storing the error produced from a preceding trial into an error memory part in the form of the control result and setting with calculation the gain of the motion control signal in proportion to the value of the error at the next trial stage. CONSTITUTION:The answer value theta' corresponding to the command value thetaof a robot 11 varies in response to the time change of the value theta. Here an error DELTAtheta is obtained by subtracting the value theta' from the value theta and stored in an error memory part 14. The error signal DELTAtheta stored in the part 14 is supplied to a gain correcting circuit 13 at the next motion control time point for calculation of the feedback gain. Thus the learning effect is obtained with reduction of the gain secured to the command value by repeating trials. This improves the motion control performance of a robot arm.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a robot arm learning control device that has a learning function for controlling the motion of a robot arm such as an industrial robot.

[Problems to be solved by conventional technology and invention] In conventional robots, even if errors occur in arm motion control,
There is no algorithm that brings about the best result by changing the output gain of the arm control signal based on the previous control result. For this reason, it was not possible to reduce the error even if the number of repetitions of the exercise was repeated.

This invention was made to improve the above-mentioned problems, and provides a robot arm learning control device that can improve control performance through repeated trials, which was previously impossible with predetermined control. The purpose is to

[Means for solving the problem and their effects, that is,
A robot arm learning control device according to the present invention includes a robot that controls arm motion based on a command value of a motion control signal, a response signal generation means that generates a response signal corresponding to the amount of motion of the robot, and a response signal generator that generates a response signal corresponding to the motion amount of the robot. error detection means for detecting an error in the arm movement by subtracting the response value of the response signal from the command value; and an error memory section for rewriting and storing the error obtained by the error detection means each time the robot performs a trial. and a gain correction circuit that corrects the gain of the next motion control signal to be supplied to the robot based on the error stored in the error memory section.

In other words, this robot arm learning control device stores the error that occurred in the previous trial as a control result in the error memory section, and calculates and sets the gain of the motion control signal in proportion to the size of the error in the next trial stage. By doing this, the error is reduced each time the trial is repeated, giving the robot a learning function.

[Embodiment] Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings.

The drawing shows its configuration, and numeral 11 in the drawing is a robot. This robot 11 has a subtracter 12 and a gain correction circuit 1.
The arm movement is attempted based on the command value θ of the movement control signal supplied via the control signal 3. The robot 11 is configured to generate a response signal in accordance with the amount of arm movement. Here, the response value is expressed as θ'. The response signals are supplied to the subtracter 12 and subtracted from the motion control signal. The output of the subtracter 12 is supplied to the gain correction circuit 13 and also to the error memory section 14 as an error signal Δθ(-0-θ'). The error memory section 14 sequentially rewrites and stores the error signal Δθ generated each time the robot 11 moves, and the stored error signal Δθ is supplied to the gain correction circuit 13 during the next robot motion control. This gain correction circuit 13 calculates and corrects the gain of the motion control signal supplied to the robot 11 based on the error signal Δθ obtained in the previous trial.

That is, the response value θ' of the robot 11 to the command value θ changes as the command value θ changes over time. The error Δθ at this time is obtained by subtracting the response value θ' from the command value θ, and is stored in the error memory section 14. The error signal Δθ stored in the error memory section 14 is supplied to the gain correction circuit 13 during the next motion control. This gain correction circuit 13 performs feedback gain calculation, and in the i+1st gain calculation, Ki+1 − + < 1 + α 1 Δθ 1
)...Given by (1). Note that Ki is the i-th gain.

In other words, in the first trial, 1Δθ1-0, so
K1-KO, that is, a fixed gain.

The second time is 2-Kl (1+α1Δθi), which is a variable gain that takes into account the previous trial results. Hereinafter, calculations are similarly made according to the gain calculation formula (1) expressed by the above recurrence formula. Note that α is an adjustment coefficient, and the control effect can be adjusted depending on the size difference. In other words, it is possible to increase the gain effect by increasing the value of α. In other words, α can be adjusted to the best value from the control effect.

Therefore, the robot arm learning control device configured as described above has a learning effect of reducing the error with respect to the command value through repeated trials, so that the control performance of the robot arm movement can be improved.

[Effects of the Invention] As detailed above, according to the present invention, the error generated in the previous trial is stored in the error memory section as a control result, and the error is moved in proportion to the size in the next trial stage. By calculating and setting the gain of the control signal, the error is reduced every m trials, and by giving the robot a learning function, it is possible to solve problems that would not be possible with conventional predetermined control. It is possible to provide a robot arm learning control device that can improve control performance by repeating possible trials.

[Brief explanation of drawings]

The drawing is a block circuit configuration diagram showing an embodiment of a robot arm learning control device according to the present invention. DESCRIPTION OF SYMBOLS 11... Robot, 12... Subtractor, 13... Gain correction circuit, 14... Error memory section, θ... Command value, θ-... Response value, Δθ... Error signal .

Claims (1)

[Claims] A robot that controls arm movement based on a command value of a motion control signal, a response signal generating means that generates a response signal corresponding to the amount of motion of the robot, and a response value of the response signal that is determined from a command value of the motion control signal. an error detection means for detecting an error in the arm movement by subtraction; an error memory section for rewriting and storing the error obtained by the error detection means for each trial of the robot; and an error stored in the error memory section. A robot arm learning control device comprising: a gain correction circuit that corrects a gain of a next motion control signal to be supplied to the robot based on the robot arm learning control device.