Buttoning Task with a Dual-Arm Robot: An Exploratory Study on a Marker-based Algorithmic Method and Marker-less Machine Learning Methods

Wakana Fujii, Kanata Suzuki, Tomoki Ando, Ai Tateishi, Hiroki Mori, Tetsuya Ogata

Abstract: In this study, we conduct the first trial ever to realize a robotic buttoning task using a dual-arm robot. A robot must handle a flexible object (clothes) and solid objects (buttons) simultaneously during the buttoning task, therefore there is no previous work due to its complexity. We design a strategy of the buttoning task using a dual-arm robot by dividing a series of motions into subtasks with the following methods: (a) a marker-based algorithmic method, markerless machine learning methods (b) without pseudo-rehearsal motions and (c) with pseudo-rehearsal motions. The pseudo-rehearsal is a consolidation learning using generated data by the firstly trained model. We examine the method (a) to make sure if the buttoning task is feasible. For real world setup, markers should not be attached to shirts. Hence, we try to achieve the first half of the task (the button-hooking task) by machine learning methods (b)(c) after we collect dataset via the robotic task with markers (a). In the experiments, we verify that pseudo-rehearsal learning contributes to adapt the different environment between the shirt with markers as the experimental setup and the shirt without markers as the daily living setup.