2022 年 40 巻 9 号 p. 817-826
In this paper, we present a learning-based method for estimating the state of cloth during putting an arm through a sleeve of shirt. To capture the dynamic change of cloth during manipulation, we take an approach that uses optical flow extracted from image streams. We adopt a deep neural network for optical flow extraction, then extend that the network to output a cloth state. To evaluate the accuracy of state estimation, we conducted two experiments: (i) putting a cylindrical cloth through an L-shaped stand, and (ii) putting a long-sleeved shirt through an arm of a doll. The experimental results indicated that our method is useful to estimate cloth state more accurately than conventional methods.