Semantic grasping for robotic manipulation through knowledge graph

Abstract

In this paper, we studied semantic grasping for robotic manipulation. For efficient robotic manipulation, the robot should change the grasping position, grasping force, and gripper type depending on the object's material, characteristics, and purpose of grasping. However, designating characteristics and grasping parameters for every object is inconvenient and inefficient. To solve this problem, we suggest a method to predict appropriate grasping parameters for objects whose grasping parameters are unknown. We show that grasping parameters can be predicted using the knowledge graph embedding method over the knowledge graph of objects' characteristics, their relation, and grasping parameters. Furthermore, we developed a method for a robot to decide the appropriate grasping parameter by collecting information about an object using its recognition module.