US 11,440,184 B2Granted

Position control device and position control method

AssigneeMitsubishi Electric Corporation

InventorsToshisada Mariyama, Mamoru Miura, Wataru Matsumoto

Life of this patent

Estimating the term…

Loading patent…

Abstract

An imaging unit, a control parameter generation unit, a control unit, and a drive unit are provided. The imaging unit captures an image including two objects. The control parameter generation unit feeds information of the captured image including the two objects into an input layer of a neural network, and outputs a position control amount for controlling the positional relation of between the two objects as an output layer of the neural network. The control unit controls current or voltage to control the positional relation between the two objects by using the outputted position control amount. The drive unit changes a position of one of the two objects by using the current or the voltage. Here, the control parameter generation unit selects the neural network from a plurality of neural networks. Therefore, even if there are differences between objects or errors in the positional relationship between the two objects, alignment can be performed more accurately.

Claims6

Dependent claims

A position control device to install a first object to a second object by causing a motor to move the first object, the device comprising:

imaging circuitry of a monocular camera to capture an image including two objects that are the first object and the second object;
and
processing circuitry programmed to utilize a plurality of neural networks, wherein each of the plurality of neural networks
has an input layer into which information of an image captured by the imaging circuitry is inputted,
has an output layer in which a control amount obtained on a basis of a learning rule is outputted, the control amount being for moving the first object in correspondence with the image inputted into the input layer, and
is trained in association with a range of control amounts to be outputted by the output layer, wherein the plurality of neural networks are trained respectively in association with a plurality of different ranges of control amounts,
wherein the processing circuitry is further programmed to
in response to the imaging circuitry capturing a first image, which includes the first object and the second object, at a first time,
determine a first range from among the different ranges of control amounts having a highest magnitude, and
select a first neural network as the neural network among the plurality of neural networks trained in association with the first range,
input information of the first image into the input layer of the first neural network,
output a first control amount, which is outputted in the output layer of the first neural network in response to the information of the first image being inputted into the input layer of the first neural network, as a control amount for causing the motor to move the first object included in the first image acquired from the imaging circuitry,
in response to the imaging circuitry capturing a second image, which includes the first object and the second object, at a second time, which is later than the first time,
determine, from among the plurality of different ranges of the control amounts, a second range within which a magnitude of the first control amount falls,
select a second neural network as the neural network among the plurality of neural networks trained in association with the second range,
input information of the second image into an input layer of the second neural network, and
output a second control amount, which is outputted in the output layer of the second neural network in response to the information of the second image being inputted into the input layer of the second neural network, as a control amount for causing the motor to move the first object included in the second image acquired from the imaging circuitry.

14 more elements

A position control method for installing a first object to a second object by causing a motor to move the first object, the method comprising:

determining a first neural network from among a plurality of neural networks, each of the plurality of neural networks including an input layer for inputting information of an image including the first object and the second object, each of the plurality of neural networks including an output layer for outputting a control amount from a corresponding range of control amounts, the first neural network being determined as the one of the plurality of neural networks whose corresponding range of control amount has a highest magnitude;
inputting information of a first image, which is captured by imaging circuitry of a monocular camera at a first time and includes the first object and the second object, into an input layer of the first neural network;
outputting a first control amount that is outputted in the output layer of the first neural network, the first control amount being a value
that is outputted as a result of the information of the first image being inputted into the input layer of the first neural network,
that is obtained on a basis of a learning rule, and
that is applied as a control amount for causing the motor to move one of the two objects included in the first image captured by the imaging circuitry;
inputting information of a second image including the first object and the second object captured by the imaging circuitry at a second time, which is later than the first time, into an input layer of a second neural network selected from among the plurality of neural networks on a basis of which of the plurality of neural networks has a corresponding range of control amounts within which a magnitude of the first control amount falls;
and
outputting a second control amount that is outputted in the output layer of the second neural network, the second control amount being a value
that is outputted as a result of the information of the second image being inputted into the input layer of the second neural network,
that is obtained on a basis of a learning rule, and
that is applied as a control amount for causing the motor to move the first object included in the second image captured by the imaging circuitry.

9 more elements

A position control method for installing a first object to a second object by causing a motor to move the first object, the method comprising:

performing a process of training a plurality of neural networks including
calculating a first movement correction amount that maximizes a defined reward on a basis of a position of the first object in information of an image that includes two objects that are the first object and the second object, and that is captured by imaging circuitry of a monocular camera at a first trial, as a first movement control amount for causing the motor to move the first object;
calculating a first recommendation value on a basis of an evaluation value indicating a degree of success based on a positional relation between the first object and the second object when the first object is moved according to the first movement control amount;
calculating, using the first recommendation value, a second movement correction amount on a basis of a position of the first object in an image captured by the imaging circuitry at a second trial, as a second movement control amount for causing the motor to move the first object;
calculating, when the imaging circuitry captures an image at a third trial, a second recommendation value on a basis of an evaluation value indicating a degree of success based on a positional relation between the first object and the second object when the first object is moved according to the second movement control amount;
calculating, using the second recommendation amount, a third movement correction amount from a position of the first object in an image captured by the imaging circuitry at the third trial, as a third movement control amount for causing the motor to move the first object;
and
selecting a neural network from among the plurality of trained neural networks, each of the plurality of neural networks including an input layer for inputting information of an image including the first object and the second object, each of the plurality of neural networks including an output layer for outputting a control amount from a corresponding range of control amounts, the selection being made on a basis of the ranges of control amounts respectively corresponding to the plurality of neural networks;
inputting information of a particular image including the first object and the second object, which is captured by the imaging circuitry, into the input layer of the selected neural network;
and
outputting a control amount that is outputted in the output layer of the selected neural network, the obtained control amount being a value
that is outputted as a result of the information of the particular image being inputted into the input layer of the selected neural network,
that is obtained on a basis of a learning rule subjected to said training, and
that is applied as a control amount for causing the motor to move the first object included in the particular image captured by the imaging circuitry.

11 more elements

Dependencies are inferred from the claim text.

Facts

Application: 16/483,381Filed 9 Feb 2017
Published: US 2020/0030970 A130 Jan 2020
Granted: 13 Sep 2022
Term ends: Not estimatedThe source records don’t support an estimate.

Classification24 codes

B25J 9/161
Characterised by the control system, structure, architecture: hardware, e.g. neural networks, fuzzy logic, interfaces, processor
B25J 9/163
Characterised by the control loop: learning, adaptive, model based, rule based expert control
B25J 9/1697
Characterised by use of sensors other than normal servo-feedback from position, speed or acceleration sensors, perception control, multi-sensor…:…
B25J 13/08
Controls for manipulators: by means of sensing devices, e.g. viewing or touching devices

Family

US 2020/0030970 A1Pre-grant publication · published 30 Jan 2020
US 11,440,184 B2This patent

From the US filing history. Foreign family members aren’t listed.

Timeline

Citations

Open elsewhere

Building an agent or pulling many patents? Use the API or MCP server instead of scraping pages. Get an API key