US 2024/0173855 A1Published application

Method for Generating Training Data for a Machine Learning (ML) Model

AssigneeSiemens Aktiengesellschaft

InventorsIngo THON, Ralf GROSS

Life of this patent

Estimating the term…

Loading patent…

Abstract

The invention relates to a method for generating training data for an ML model. The training data is designed to configure the ML model using a machine learning method. In particular, the ML model is designed to be used as part of a method for ascertaining control data for a gripping device for gripping an object. The invention is characterized by the steps of: —selecting an object, —selecting starting data of the object above a flat surface, —generating a falling movement of the object in the direction of the flat surface beginning with the starting data, —capturing an image of the object after the movement of the object has come to a standstill on the flat surface, —assigning an identifier to the captured image, said identifier comprising ID information for a stable position assumed by the object, wherein the stable position assumed by the object is designed and configured such that all of the object position data that can be converted into one another by means of a movement and/or rotation about a surface normal of the flat surface is assigned to the assumed stable position, and—storing the training data comprising the captured image and the identifier assigned thereto.

Claims8

Dependent claims

A method for generating training data for an ML model (142, 162 ,

wherein the training data are designed and configured to configure the ML model (142, 162) by using a machine learning method,
and wherein in particular the ML model (142, 162) is designed and configured to be used as part of a method for ascertaining control data for a gripping device (120, 122) for gripping an item (200, 210, 220, 300, 310),
characterized by the method steps of:
selecting an item (200, 210, 220, 300, 310),
selecting starting data relating to the item (200, 210, 220, 300, 310) above a planar surface (112),
producing a falling motion for the item (200, 210, 220, 300, 310) in the direction of the planar surface (112) beginning with the starting data,
capturing an image (132) of the item (200, 210, 220, 300, 310) after a motion of the item (200, 210, 220, 300, 310) has stopped on the planar surface (112),
assigning an identifier to the captured image (132), the identifier comprising ID information for a stable attitude adopted by the item (200, 210, 220, 300, 310), the stable attitude adopted by the item being designed and configured in such a way that all of those attitude data relating to the item that can be translated into one another by way of a shift and/or a rotation about a surface normal of the planar surface are assigned to this adopted stable attitude,
storing the training data comprising the captured image and the identifier assigned to said image.

6 more elements

A method for generating training data for an ML model (142, 162),

wherein the training data are designed and configured to configure the ML model (142, 162) by using a machine learning method,
and wherein in particular the ML model (142, 162) is designed and configured to be used as part of a method for ascertaining control data for a gripping device (120, 122) for gripping an item (200, 210, 220, 300, 310),
characterized by the method steps of
selecting a 3D model of an item (250, 350),
selecting starting data relating to the 3D model of the item (250, 350) above a virtual planar surface,
simulating a falling motion for the 3D model of the item (250, 350) in the direction of the virtual planar surface beginning with the starting data,
creating an image (132) of the 3D model of the item (250, 350) after the simulated motion of the 3D model of the item (250, 350) has stopped on the virtual planar surface,
assigning an identifier to the created image (132), the identifier comprising ID information for a stable attitude adopted by the 3D model of the item (250, 350), the stable attitude adopted by the item being designed and configured in such a way that all of those attitude data relating to the item that can be translated into one another by way of a shift and/or a rotation about a surface normal of the virtual planar surface are assigned to this adopted stable attitude,
storing the training data comprising the created image and the identifier assigned to said image.

6 more elements

Dependencies are inferred from the claim text.

Facts

Application: 18/283,228Filed 4 Mar 2022
Priority: EP 2116393322 Mar 2021
Published: 30 May 2024
Term ends: Not estimatedThe source records don’t support an estimate.

Classification12 codes

B25J 9/1612
Program controls: characterised by the hand, wrist, grip control
G06F 30/27
Design optimisation, verification or simulation: using machine learning, e.g. artificial intelligence, neural networks, support vector machines…
B25J 9/163
Characterised by the control loop: learning, adaptive, model based, rule based expert control
G06N 3/0455
Combinations of networks: auto-encoder networks; Encoder-decoder networks

Family

No related US applications are recorded in the indexed filing history.

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