Data Augmentation for FPCB Picking in Heavy Clutter via Image Blending

In this paper, we propose a picking solution on flexible printed circuit board (FPCB) which is a key electronic component for flexible smartphones and wearable devices. We provide an analysis on FPCBs to indicate where a vacuum gripper can pick. From the analysis, we propose a FPCB picking system. Since FPCBs are commonly small, thin, flat, and they are highly overlapped each others in heavily cluttered environment, a manual labeling process may take huge resources or even impossible. Therefore, we propose an effective method to generate training dataset, and provide experiments of picking performance on generated dataset. Experiment verified that a grasp-pose detection network trained with our generated dataset is good enough to detect picking region on FPCBs.