Grasping Position Detection Using Template Matching and Differential Evolution for Bulk Bolts

Abstract

In order to pick bolts stacked in bulk using a robot, it is necessary to obtain the most suitable grasping position. Recently, deep learning-based methods have been proposed. However, learning takes a long time and much effort to prepare the training. In order to avoid this, this research proposes a template matching-based method. First of all, we defined some states where the bolt can be grasped stably. By efficiently searching the grasping position where the objective function is maximized using differential evolution (DE), the proposed method can acquire the grasping position in two seconds. After an original gripper was made using a 3D printer and attached to the robot, we experimented using 60 bolts (M6-20). The results show 86% of the success rate.