| Autor: |
Daisuke TANAKA, Solvi ARNOLD, Kimitoshi YAMAZAKI |
| Jazyk: |
japonština |
| Rok vydání: |
2018 |
| Předmět: |
|
| Zdroj: |
Nihon Kikai Gakkai ronbunshu, Vol 84, Iss 864, Pp 18-00069-18-00069 (2018) |
| Druh dokumentu: |
article |
| ISSN: |
2187-9761 |
| DOI: |
10.1299/transjsme.18-00069 |
| Popis: |
In this paper, we describe a motion planning method for cloth manipulation. We address the problem that only the current shape of the cloth product and its target shape are given. To accomplish this task, it is necessary to determine how to manipulate the cloth and to predict the manipulated shape of the cloth. Therefore, we propose a novel method using deep neural network. An input of the network is a voxelized cloth shape and conceivable manipulations, and an output is a voxelized cloth shape after the manipulations. The effectiveness of the proposed method was proven by experiments using a dual-armed robot. It included quantitative evaluation about the manipulated shape of a rectangular cloth. It was also studied that how to prepare appropriate learning data and do learning well. |
| Databáze: |
Directory of Open Access Journals |
| Externí odkaz: |
|
