Hanging Work of T-Shirt in Consideration of Deformability and Stretchability * Yosuke Koishihara, Solvi Arnold, Kimitoshi Yamazaki Faculty of Engineering Shinshu University Nagano, Nagano, Japan kyamazaki@shinshu-u.ac.jp Takamitsu Matsubara Graduate School of Information Science Nara Institute of Science and Technology(NAIST) Ikoma, Nara, Japan takam-m@is.naist.jp Abstract— This paper describes a hanging work, which is one of the work related to cloth products. By using deforma- bility and stretchability of clothes successfully, it is possible to efficiently proceed with a hanging work. We focus on cloth wrinkles and superimposed relation of objects, and propose a method to recognize the progress of hanging work on-line. In a detection processing of wrinkles, after using edge detection filters, clustering is applied, and the direction and inclination of the wrinkle are extracted. By statistically observing the large number of wrinkles, we can estimate the current tension state of cloth. The effectiveness of the proposed method was proven through experiments using a dual-armed robot. Index Terms— Hanging work, wrinkles, deformability and strechability I. I NTRODUCTION Cloth product is an important tool supporting human social life. Therefore, there are also many work related to cloth product in daily life. Many of them are nonproductive, such as cleaning and drying, therefore it is considered useful if they can be replaced by automated machines. In this paper, we focus on hanging clothes, which is one of the work related to cloth products. In this work, it is necessary to consider not only the deformability of the cloth but also the fact that the cloth streches. Especially, stretchability is helpful characteristic that makes it possible to widen a hole originally small. By using it successfully, it is possible to efficiently proceed with a hanging work. We can find studies on automatic operation of cloth products [1-4]. Rectangular cloth products and clothing were used, and their folding or unfolding tasks have been targetted. Cu´ en-Roch´ ın et al. [5] proposed an action selection method for handling a planar cloth. The recognition method is based on matching between a 3D point cloud and a physical model, and the result is used to spread a square cloth. As actions that fold cloth products has also been noted, it is possible to transfer cloth products that are placed casually into the desired form. Kita et al. [6] used a 3D deformable model, and obtained a correspondence between the model and an input pointcloud that was captured by a trinocular stereo camera. Doumanoglou et al. [7] succeeded in recognizing the type * This work was partly supported by NEDO and JSPS KAKENHI Grant Numbers 26700024. and shape of clothing items during the unfolding process, using a 3D range camera. Their framework also provided the next grasping point for subsequent manipulation. Stria et al. [8] unfolded a clothing item by means of shape estimation using a polygonal model. As their main manipulation was pick-and-place, they designed recognition functions to find feasible grasping points for the next step in the unfolding process. In these studies, even though the authors assumed the de- formation of the cloth, but they did not assume strechability. Therefore, when considering the work of inserting something into an empty hole in a cloth product, if the size of the insert is larger than the hole not pulled, it is judged that the insertion work can not be performed. However, many cloth products are elastic materials, and depending on how they are handled, the above-described insertion work is also possible. In this study, we aim to automate such work, and focus on hanging work as an example. Besides this, we consider that there is similar development such as assisting work of clothes wearing, packing work at logistics sites, assembling work at the manufacturing site. In this paper, we combine two kinds of state recognition methods; one is a method of confirming that a hanger can be moved to a desired arrangement with respect to the cloth, and another is a method of estimating the tension state of the cloth. Also consideration is given to high-speed processing, making it possible to observe the state of tension during hanger manipulation. The paper is organized as follows: Section II presents related work. Section III explaines our approach. Section IV and V explains the proposed method. Section VI shows ex- perimental results, and Section VII presents the conclusions of this paper. II. RELATED WORK It is important to observe the tension state of cloth in the hanging work. In that sense, information processing about wrinkles is an important element. Even in automatic opera- tion of cloth products in intelligent robotics, there are several studies focusing on the state of wrinkles. Yamazaki and Inaba [9] devised a wrinkle-specific image feature expression and proposed a method to detect clothing captured in daily