Machine learning method and robot system

Patent metadata

Field	Value
Publication number	US-12552019-B2
Application number	US-202017995673-A
Country	US
Kind code	B2
Filing date	Apr 28, 2020
Priority date	Apr 28, 2020
Publication date	Feb 17, 2026
Grant date	Feb 17, 2026

Abstract

A machine learning method for learning an action of a robot including a hand to pick out a workpiece from a container containing a plurality of the workpieces stacked in bulk and install the workpiece such that the workpiece is in a predetermined installation state includes learning a reverse-order action of removing, by the hand, the workpiece in the predetermined installation state after completion of installation, and learning an installation order of the workpiece based on a learning result of the reverse-order action of removing the workpiece.

First claim

What is claimed is:

1 . A machine learning method for learning an action of a robot including a hand to pick out two or more workpieces from a container containing a plurality of the workpieces stacked in bulk and install the two or more workpieces such that the two or more workpieces are in a predetermined installation state, the machine learning method comprising:
learning a reverse-order action of removing, by the hand, the two or more workpieces in the predetermined installation state after completion of installation; and learning an installation order of the two or more workpieces based on a learning result of the reverse-order action of removing the two or more workpieces, wherein the learning of the reverse-order action of removing the two or more workpieces includes repeating the reverse-order action of removing the two or more workpieces by the hand until all or at least one, but not all, of the two or more workpieces in the predetermined installation state is removed, the learning of the reverse-order action of removing the two or more workpieces further includes changing an order of removing each of the two or more workpieces in the predetermined installation state and repeating the reverse-order action by the hand according to the changed order until all or at least one, but not all, of the two or more workpieces in the predetermined installation state is removed, and the learning of the reverse-order action of removing the two or more workpieces includes learning a plurality of reverse-order actions in which the order of removing each of the two or more workpieces is changed to be different from each other.

2 . The machine learning method according to claim 1 , wherein the learning of the installation order of the two or more workpieces includes learning both a plurality of installation orders and priorities of the plurality of installation orders based on the learning result of the reverse-order action of removing the two or more workpieces.

3 . The machine learning method according to claim 1 , wherein the learning of the reverse-order action of removing the two or more workpieces includes repeating the reverse-order action of removing the two or more workpieces by the hand until all or at least one, but not all, of the two or more workpieces in the predetermined installation state is removed in a state in which a portion of the order of removing each of the two or more workpieces has been set.

4 . The machine learning method according to claim 1 , further comprising:
learning a holding position of each of the two or more workpieces held by the hand based on the learning result of the reverse-order action of removing the two or more workpieces.

5 . The machine learning method according to claim 4 , wherein the learning of the holding position of each of the two or more workpieces includes learning the holding position of each of the two or more workpieces with consideration of a constraint condition including at least one of a holding prohibited portion of each of the two or more workpieces or an obstacle in a vicinity of each of the two or more workpieces.

6 . The machine learning method according to claim 1 , further comprising:
notifying a user of a review of at least one of the hand or a jig pallet on which the two or more workpieces is installed, based on a success probability of the reverse-order action of removing the two or more workpieces.

7 . The machine learning method according to claim 1 , wherein the learning of the reverse-order action of removing the two or more workpieces includes learning a reverse-order action of removing a plurality of types of the workpieces by the hand.

8 . A robot system comprising:
a robot including a hand to pick out two or more workpieces from a container containing a plurality of the workpieces stacked in bulk and install the two or more workpieces such that the two or more workpieces are in a predetermined installation state; a machine learning device configured to learn an action of the robot; and a controller configured or programmed to control the action of the robot based on a learning result of the machine learning device; wherein the machine learning device is configured to learn a reverse-order action of removing, by the hand, the two or more workpieces in the predetermined installation state after completion of installation, and learn an installation order of the two or more workpieces based on a learning result of the reverse-order action of removing the two or more workpieces, the machine learning device is configured to repeat the reverse-order action of removing the two or more workpieces by the hand until all or at least one, but not all, of the two or more workpieces in the predetermined installation state is removed, the machine learning device is configured to change an order of removing each of the two or more workpieces in the predetermined installation state and repeat the reverse-order action by the hand according to the changed order until all or at least one, but not all, of the two or more workpieces in the predetermined installation state is removed, and the machine learning device is configured to learn the reverse-order action of removing the two or more workpieces by learning a plurality of reverse-order actions in which the order of removing each of the two or more workpieces is changed to be different from each other.

9 . The robot system according to claim 8 , wherein the machine learning device is configured to learn a holding position of each of the two or more workpieces held by the hand based on the learning result of the reverse-order action of removing the two or more workpieces.

10 . The robot system according to claim 9 , wherein the machine learning device is configured to select the holding position of each of the two or more workpieces that allows the two or more workpieces to be picked out from the container and allows the two or more workpieces to be installed.

11 . The robot system according to claim 10 , wherein the machine learning device is configured to extract the holding position of each of the two or more workpieces that allows the two or more workpieces to be picked out from the container based on a success probability of picking out the two or more workpieces, and extract the holding position of each of the two or more workpieces that allows the two or more workpieces to be installed based on a success probability of installing the workpiece.

12 . The machine learning method according to claim 1 , wherein the learning of the reverse-order action of removing the two or more workpieces includes repeating the reverse-order action of removing the two or more workpieces by the hand until all or at least one, but not all, of the two or more workpieces in the predetermined installation state is removed in a state in which a portion of the order of removing each of the two or more workpieces has been set.

13 . The machine learning method according to claim 2 , wherein the learning of the reverse-order action of removing the two or more workpieces includes repeating the reverse-order action of removing the two or more workpieces by the hand until all or at least one, but not all, of the two or more workpieces in the predetermined installation state is removed in a state in which a portion of the order of removing each of the two or more workpieces has been set.

14 . The machine learning method according to claim 1 , further comprising:
learning a holding position of each of the two or more workpieces held by the hand based on the learning result of the reverse-order action of removing the two or more workp

Assignees

• Yamaha Motor Co Ltd

Inventors

• Urata Satoshi

Classifications

• B25J9/163 Primary

  learning, adaptive, model based, rule based expert control · CPC title

• G06N20/00

  Machine learning · CPC title

• G05B2219/39543

  Recognize object and plan hand shapes in grasping movements · CPC title

• G05B2219/40543

  Identification and location, position of components, objects · CPC title

• B25J9/1612 Primary

  characterised by the hand, wrist, grip control · CPC title