Category Classification of Deformable Object using Hybrid Dynamic Model for Robotic Grasping

Abstract

This work studies the problem of classification of a hung garment in the unfolding procedure by a home service robot. The sheer number of unpredictable configurations that the deformable object can end up in makes the visual identification of the object shape and size difficult. In this paper, we propose a hybrid dynamic model to recognize the pose of hung garment using a single manipulator. A dataset of hung garment is generated by capturing the depth images of real garments at the robotic platform (real images) and also the images of garment mesh model from offline simulation (synthetic images) respectively. Deep convolutional neural network is implemented to classify the category and estimate the pose of garment. Experiment results show that the proposed method performs well and is applicable to different garments in robotic manipulation.