Yamakawa Laboratory, The University of Tokyo

東京大学　山川研究室

This study introduces a novel method for selecting suspension points on randomly positioned towel-like objects to facilitate a series of sequential grasping, spreading, and alignment operations. Unlike traditional approaches that primarily focus on successful object retrieval, our method prioritizes the state of the suspended towel-like object, aiming to ensure that only one clear corner appears at the bottom during the suspension process to facilitate the aligning process. Semantic segmentation plays a crucial role in detecting corners and edges on randomly placed towel-like objects, from which we identify optimal suspension points based on various features, including area, predicted probability, depth, and distance from the object’s contour. To fulfill our primary goal of guaranteeing the presence of a clear corner at the lowest point of a suspended towel-like object, we introduce a specially designed Multi-Time-Step filter and image-sharpening technique to enhance the stability and accuracy of corner and edge detection. Through a series of experiments conducted on randomly placed towel-like objects, our proposed method successfully achieved its objective by significantly increasing the success rate of corner grasping when compared to random suspension point selection.

• Taohan Wang, Mamoru Oka, Kenichi Murakami, Shouren Huang, Hirofumi Sumi, Masatoshi Ishikawa and Yuji Yamakawa: Robot system for sequential suspending point selection, grasping, spreading, and aligning manipulation of randomly placed towel-like objects, Robotics and Autonomous Systems, 105053 (2025) [link]