Key points are not available for this paper at this time.
The physical properties of highly deformable objects such as clothing poses a challenging problem for autonomously acting systems. Especially, grasping and manipulation require new approaches that can accommodate for an object's variable and changing appearance. In this paper, we present a system that is capable of fully autonomously transforming a clothing item from a random crumpled configuration into a folded state. We describe a method to compute valid grasp poses on the cloth which accounts for deformability. Our algorithm includes a novel fold detection and grasp generation strategy, which suggests grasp poses on cloth folds. Machine learning techniques are used to evaluate these grasp poses. In our experiments, we use a stock PR2 robot whose two arms alternatingly perform grasps on a T-shirt equipped with fiducial markers. The goal of this grasp sequence is to bring the T-shirt into a configuration from which the robot can fold it. In several experiments, we demonstrate the performance of our approach.
Bersch et al. (Thu,) studied this question.