Learning dataset generation device and learning dataset generation method

{
  "publication_number": "US12217489B2",
  "country": "US",
  "kind": "B2",
  "title": "Learning dataset generation device and learning dataset generation method",
  "abstract": "A learning dataset generation device includes: a memory that stores three-dimensional CAD data of a workpiece and a container; and one or more processors including hardware, wherein the one or more processors are configured to use the three-dimensional CAD data of the workpiece and the container, stored in the memory, to generate, in a three-dimensional virtual space, a plurality of imaging objects in which a plurality of the workpieces are bulk-loaded in different forms inside the container, acquire a plurality of virtual distance images by measuring each of the generated imaging objects by means of a virtual three-dimensional measurement machine disposed in the three-dimensional virtual space, accept at least one teaching position for each of the acquired virtual distance images, and generate a learning dataset by associating the accepted teaching position with each of the virtual distance images.",
  "claims": [
    "1. A learning dataset generation device comprising: a memory that stores three-dimensional CAD data of a workpiece and a container; and one or more processors comprising hardware, wherein the one or more processors are configured to: use the three-dimensional CAD data of the workpiece and the container, stored in the memory, to generate, in a three-dimensional virtual space, a plurality of imaging objects in which a plurality of the workpieces are bulk-loaded in different forms inside the container, acquire a plurality of virtual distance images by measuring each of the generated imaging objects by means of a virtual three-dimensional measurement machine disposed in the three-dimensional virtual space, accept at least one teaching position for each of the acquired virtual distance images, generate a learning dataset by associating the accepted teaching position with each of the virtual distance images, and use a distance image of the container in a three-dimensional real space, which is acquired by means of a three-dimensional measurement machine installed in the three-dimensional real space, to set the position of the virtual three-dimensional measurement machine relative to the container in the three-dimensional virtual space to a position matching the arrangement of the three-dimensional measurement machine relative to the container.",
    "2. A learning dataset generation method comprising: using three-dimensional CAD data of a workpiece and a container to generate, in a three-dimensional virtual space, a plurality of imaging objects in which a plurality of the workpieces are bulk-loaded in different forms inside the container; acquiring a plurality of virtual distance images by measuring each of the generated imaging objects by means of a virtual three-dimensional measurement machine disposed in the three-dimensional virtual space; accepting at least one teaching position for each of the acquired virtual distance images; generating a learning dataset by associating the accepted teaching position with each of the virtual distance images; and using a distance image of the container in a three-dimensional real space, which is acquired by means of a three-dimensional measurement machine installed in the three-dimensional real space, to set the position of the virtual three-dimensional measurement machine relative to the container in the three-dimensional virtual space to a position matching the arrangement of the three-dimensional measurement machine relative to the container."
  ],
  "description_excerpt": "The present disclosure relates to a learning dataset generation device and a learning dataset generation method.\n\nThere is a known method of generating a dataset to be used in machine learning by acquiring distance images of a plurality of workpieces by means of a three-dimensional measurement machine, and by storing teaching data comprising the acquired distance images in association with a label map indicating a teaching position (for example, see Japanese Unexamined Patent Application, Publication No. 2019-58960).\n\nThe method in Japanese Unexamined Patent Application, Publication No. 2019-58960 is used, for example, for generating a learned model that estimates a take-out position when a plurality of workpieces bulk-loaded in a container are taken out one by one with a hand attached to a robot. In order to generate a highly precise learned model, it is necessary to prepare a huge number of datasets. In other words, every time a dataset is generated, it is necessary to bulk-load the plurality of workpieces in a different form and acquire distance images by means of the three-dimensional measurement machine.",
  "cpc": [
    "G06V 10/7747",
    "B25J 9/163",
    "B25J 9/1656",
    "B25J 9/1697",
    "G05B 2219/40053",
    "G05B 2219/45063",
    "G06F 30/12",
    "G06T 7/70"
  ],
  "ipc": [
    "G06F 30/12",
    "G06T 7/70",
    "G06V 10/774"
  ],
  "assignees": [
    "Fanuc Corp"
  ],
  "inventors": [
    "Toshiyuki Ando"
  ],
  "filing_date": "2021-02-17",
  "publication_date": "2025-02-04",
  "grant_date": "2025-02-04",
  "priority_date": "2020-02-19",
  "application_number": "US-202117792550-A",
  "family_id": "77391305",
  "cited_by_count": 0,
  "citations": [
    "EP1256860A2",
    "US20020169522A1",
    "JP2002331480A",
    "US20110010009A1",
    "WO2015140922A1",
    "US20170124120A1",
    "US20180130376A1",
    "US20180276501A1",
    "JP2018161692A",
    "WO2019055848A1",
    "US20190084151A1",
    "US20190091869A1",
    "JP2019058960A",
    "WO2019167300A1",
    "JP2019153863A",
    "US20200410754A1",
    "JP2019153246A",
    "WO2019172101A1",
    "US20210049033A1",
    "US20220317647A1"
  ]
}