Grasping complex shapes with the integration of high-speed vision and machine learning in a dynamic situation

Abstract

It is necessary to grasp a complex object with a certain level of speed and accuracy to substitute humans with robots in some studies. A system using machine learning can recognize complex objects, but its computation cost is so high such that it cannot grasp a moving object. In contrast, a system using high-frame-rate (HFR) vision can recognize objects quickly but cannot recognize complex shapes. In this study, we constructed a system combining HFR vision and machine learning to compensate for each other and attempt high-speed grasping for objects with complex shapes.