Grasp plannind under task-specific contact constraints

Abstract

Several aspects have to be addressed before realizing the dream of a robotic hand-arm system with human-like capabilities, ranging from the consolidation of a proper mechatronic design, to the development of precise, lightweight sensors and actuators, to the efficient planning and control of the articular forces and motions required for interaction with the environment. This thesis provides solution algorithms for a main problem within the latter aspect, known as the {\em grasp planning} problem: Given a robotic system formed by a multifinger hand attached to an arm, and an object to be grasped, both with a known geometry and location in 3-space, determine how the hand-arm system should be moved without colliding with itself or with the environment, in order to firmly grasp the object in a suitable way.

Central to our algorithms is the explicit consideration of a given set of hand-object contact constraints to be satisfied in the final grasp configuration, imposed by the particular manipulation task to be performed with the object. This is a distinguishing feature from other grasp planning algorithms given in the literature, where a means of ensuring precise hand-object contact locations in the resulting grasp is usually not provided. These conventional algorithms are fast, and nicely suited for planning grasps for pick-an-place operations with the object, but not for planning grasps required for a specific manipulation of the object, like those necessary for holding a pen, a pair of scissors, or a jeweler's screwdriver, for instance, when writing, cutting a paper, or turning a screw, respectively. To be able to generate such highly-selective grasps, we assume that a number of surface regions on the hand are to be placed in contact with a number of corresponding regions on the object, and enforce the fulfilment of such constraints on the obtained solutions from the very beginning, in addition to the usual constraints of grasp restrainability, manipulability and collision avoidance.

The proposed algorithms can be applied to robotic hands of arbitrary structure, possibly considering compliance in the joints and the contacts if desired, and they can accommodate general patch-patch contact constraints, instead of more restrictive contact types occasionally considered in the literature. It is worth noting, also, that while common force-closure or manipulability indices are used to asses the quality of grasps, no particular assumption is made on the mathematical properties of the quality index to be used, so that any quality criterion can be accommodated in principle. The algorithms have been tested and validated on numerous situations involving real mechanical hands and typical objects, and find applications in classical or emerging contexts like service robotics, telemedicine, space exploration, prosthetics, manipulation in hazardous environments, or human-robot interaction in general.