Machine learning device and machine learning method for learning optimal object grasp route

Patent metadata

Field	Value
Publication number	US-10692018-B2
Application number	US-201715708130-A
Country	US
Kind code	B2
Filing date	Sep 19, 2017
Priority date	Sep 27, 2016
Publication date	Jun 23, 2020
Grant date	Jun 23, 2020

Abstract

A machine learning device according to the present invention learns an operation condition of a robot that stores a plurality of objects disposed on a carrier device in a container using a hand for grasping the objects. The machine learning device includes a state observation unit for observing the positions and postures of the objects and a state variable including at least one of cycle time to store the objects in the container and torque and vibration occurring when the robot grasps the objects during operation of the robot; a determination data obtaining unit for obtaining determination data for determining a margin of each of the cycle time, the torque, and the vibration against an allowance value; and a learning unit for learning the operation condition of the robot in accordance with a training data set constituted of a combination of the state variable and the determination data.

First claim

What is claimed is:

1. A machine learning device for learning an operation condition of a robot that stores a plurality of objects disposed on a carrier device in a container using a hand for grasping the plurality of objects, the machine learning device comprising:
a hardware processor configured to
observe positions and postures of the plurality of objects and a state variable including cycle time to store the plurality of objects in the container and torque and vibration occurring when the robot grasps the plurality of objects, during operation of the robot,
obtain determination data for determining a margin of each of the cycle time, the torque, and the vibration against a respective allowance value,
learn the operation condition of the robot in accordance with a training data set constituted of a combination of the state variable and the determination data, and
in response to the cycle time, the torque, and the vibration being equal to or less than the allowance values, cause the hand to grasp the plurality of objects in a grasp order to minimize the cycle time.

2. The machine learning device according to claim 1 , wherein the cycle time is a time from when the robot begins storing the plurality of objects in the container until the robot completes storing the plurality of objects in the container.

3. The machine learning device according to claim 1 , wherein the torque is calculated based on a current flowing through a motor for driving the robot.

4. The machine learning device according to claim 1 , wherein the vibration is calculated based on an acceleration detected by an acceleration sensor provided in the hand.

5. The machine learning device according to claim 1 , the hardware processor is configured to determine the grasp order of the plurality of objects based on a learning result in accordance with the training data set.

6. The machine learning device according to claim 1 , wherein the machine learning device is connected to the robot through a network, and
the hardware processor is configured to obtain a present state variable through the network.

7. The machine learning device according to claim 1 , wherein the machine learning device is installed in a cloud server.

8. A machine learning device, for learning an operation condition of a robot that stores a plurality of objects disposed on a carrier device in a container using a hand for grasping the plurality of objects, the machine learning device comprising:
a hardware processor configured to
observe positions and postures of the plurality of objects and a state variable including cycle time to store the plurality of objects in the container and torque and vibration occurring when the robot grasps the plurality of objects, during operation of the robot,
obtain determination data for determining a margin of each of the cycle time, the torque, and the vibration against a respective allowance value,
learn the operation condition of the robot in accordance with a training data set constituted of a combination of the state variable and the determination data,
calculate a reward based on the determination data,
update a value function used for estimating a grasp order of the plurality of objects so as to reduce at least one of the cycle time, the torque, or the vibration, based on the reward,
update an action-value table that corresponds to the grasp order of the plurality of objects, based on the state variable of the cycle time, the torque, and the vibration and the reward, and
update an action-value table that corresponds to at least one of the cycle time, the torque, or the vibration when another robot having a same configuration as the robot stores other objects in a container, based on a state variable of the another robot and the reward.

9. The machine learning device according to claim 8 , wherein the hardware processor is configured to calculate the reward based on at least one of the cycle time, the torque, or the vibration.

10. A machine learning method for learning an operation condition of a robot that stores a plurality of objects disposed on a carrier device in a container using a hand for grasping the plurality of objects, the machine learning method comprising the steps of:
observing positions and postures of the plurality of objects and a state variable including cycle time to store the plurality of objects in the container and torque and vibration occurring when the robot grasps the plurality of objects, during operation of the robot;
obtaining determination data for determining a margin of each of the cycle time, the torque, and the vibration against a respective allowance value;
learning the operation condition of the robot in accordance with a training data set constituted of a combination of the state variable and the determination data; and
in response to the cycle time, the torque, and the vibration being equal to or less than the allowance values, causing the hand to grasp the plurality of objects in a grasp order to minimize the cycle time.

Assignees

• Fanuc Corp

Inventors

• Ooba Masafumi

Classifications

• B25J9/163 Primary

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

• G06N20/00 Primary

  Machine learning · CPC title

• Y02P90/02

  Total factory control, e.g. smart factories, flexible manufacturing systems [FMS] or integrated manufacturing systems [IMS] · CPC title

• G05B2219/40007

  Optimize sequence of pick and place operations upon arrival of workpiece on conveyor · CPC title

• G05B2219/39106

  Conveyor, pick up article, object from conveyor, bring to test unit, place it · CPC title