Design, creation, and validation of a comprehensive database infrastructure for robotic grasping

This paper describes the design, creation, and validation of an infrastructure for grasping. The infrastructure is premised on the ability to use large amounts of memory to store grasping algorithms in a database that can be applied broadly with any hand on any arm. This effort includes the development of a software architecture; a language syntax for configuration, shape-matching, and algorithms for grasping; and software for organizing grasp selection, configuring initial template grasps through human supervision, refining grasp placement, and refining grasp forces. An accurate dynamic simulation of a variety of robotic manipulators, hands, and objects was used to validate the approach, with positive results. The presented database approach supports generic algorithms for both fingertip and whole hand grasping of a variety of object types. Fingertip and whole hand grasps are also known as precision grasp and power grasp respectively in some literature, for example [1]. Algorithms based on grasp templates created by human supervisors and refined through automated refinement work effectively and efficiently. The presented techniques work with several types of force and touch sensors and with a variety of object shapes and physical consistencies.