A General Robotic Framework for Automated Cloth Assembly

In the pipeline of garment manufacturing, the assembly of cloth pieces using fixtures is a widely used technique to reduce the reliance on skilled workers and to improve the sewing quality. The assembly is achieved by aligning the holes on the cloth pieces with a set of vertical locating pins. Such a task, however, is difficult to be automatized by robots due to the complex dynamics of the highly deformable cloth material. In this paper, we present a general visual-based approach to automatically align cloth pieces with target pins without any prior knowledge. Given a piece of cloth, the system will detect and track the holes on it and compute the position of each hole in the 3D space. After exploring the local features of the cloth piece by moving the robotic arm several times, the system can learn the deformation model of the cloth piece and then generate an appropriate control policy which maps perception feedbacks to the control velocity of the robotic end-effectors. We demonstrate the performance of our system on an IRB 14000 YuMi robot with an Intel Realsense RGB-D camera mounted. Experiments show that our system can robustly accomplish the assembly task for cloth pieces with different materials and different patterns of locating holes.