Research on Object Category and Grasping Control for Housework Assisting Robot

Abstract

In this paper, an object category method and grasping method is proposed for housework assisting robot. Multiple kinds of objects that are trained previously by CNN (Conventional Neural Network) can be recognized and categorized by the recognized class names. Other unknown objects are categorized by the similarity of object features. The recognized objects of same category are grasped by the robot and moved to the specified places. The grasping postures are decided by the detected spatial shapes of objects.