Author: Shenli Yuan

Publisher:

Published: 2022

Total Pages:

ISBN-13:

This work describes the development of a class of dexterous robot hands that use steerable, continuously rotating fingertips to perform complex in-hand manipulation of grasped objects, a task that has alluded many widely used robot graspers. While a typical laboratory robot with a parallel-jaw gripper can perform basic pick-and-place tasks, they lack the fine manipulation proficiency needed for more demanding, precise and varied tasks. Inspired by the human hand, many research efforts are focused on creating and controlling human-like (anthropomorphic) hands in the hopes of duplicating human dexterity. One goal of these designs has been to enable robotic in-hand manipulation. Except for small motions, in-hand manipulation with these hands requires grasp-gaiting, a complex process where the fingers "walk" across the object, making and breaking contacts in order to propel the object to a desired pose, which can be an inefficient and difficult approach for a variety of tasks. So far, the fragility, cost, and complexity of these devices have precluded use outside of a laboratory setting. While we share with other researchers the goal of enabling in-hand manipulation, we achieve it by completely different means. We have developed a series of highly non-anthropomorphic grasping devices - the Roller Graspers - that use rotating fingertips to perform full six-degrees-of-freedom (DoF) in-hand manipulation on grasped objects. We do this by intelligently driving the fingertips across the object. The first Roller Grasper used steerable cylindrical fingertip rollers mounted on a pivoting linkage, for a total of three DoF for each of its three fingers. Using scripted motion control, it demonstrated the feasibility of our in-hand manipulation concept. The second and third versions used spherical fingertips which afforded better grasp stability and range of motion. These designs also supported our development of a hierarchical manipulation architecture that allowed the roller graspers to achieve autonomous in-hand manipulation. The manipulation architecture consisted of a sample-based high-level planner and a heuristic low-level policy that allowed the grasper to perform full 6-DoF manipulation of objects with a variety of shapes and sizes. The final version of our hand, called Tactile-Enabled Roller Grasper (TERG), incorporated a novel tactile sensing system that could extract the surface contour of a grasped object as well as the shear force applied at the contact location, even while the fingertips were rotating. This enabled more diverse and robust in-hand manipulation that was not possible in the previous generations.