Analysis of the inertia and dynamics of grasped objects, for choosing optimal grasps to enable torque-efficient post-grasp manipulations

This paper addresses the problem of choosing between several different possible grasps on an object, in order to enable efficient manipulation of that object after it has been grasped. In this work, we assume mass distribution of the object, object's inertia tensor, and robot's dynamic model are known a priori. We then show how each possible grasp can be expressed as an augmented dynamics model, which combines the inertias of both the robot and the grasped object in a single formulation. The augmented dynamics model can then be used to estimate the joint torques needed to move the object along a desired post-grasp trajectory. We propose a cost function based on these joint torques, and show how this cost can be conveniently used to choose the best grasp for achieving post-grasp trajectories with minimal effort. We present the results of experiments using a 7-DOF redundant manipulator, grasping objects of different shapes. Firstly, we present both simulations and empirical measurements on a real robot, which show that different grasp positions on the same object do indeed lead to significantly different amounts of effort to move that object along a post-grasp trajectory. Secondly, we evaluate how accurately the augmented dynamics formulation, combined with a physics simulator, is able to predict post-grasp joint torques, by comparing predicted torques against torques measured empirically during task execution on a real robot. We show that, even though predicted and measured torques are not identical, they are sufficiently well correlated to enable correct selection of the best grasp to enable the least-effort post-grasp motion. Finally, we demonstrate our grasp selection strategy on objects of two different shapes, and show that our method successfully chooses the best grasp in both cases.