Categorizing arbitrarily shaped objects based on grasping configurations

In this paper, we investigate categorizing arbitrarily shaped objects on the basis of how they are grasped. Using a database that consists of 28 objects, we conducted 1200 grasping experiments in a specifically developed simulation environment. The results show that, using a generic representation of grasping rules, objects can be categorized into parent–child relationships that allow grasping rules for “parent” objects to be used successfully for “child” objects. This categorization is independent of the geometric features of objects and only dependent on the grasping configuration for each object. These results could provide an important step toward developing user-adaptive robots.