A comparative study of many randomized algorithms to calibrate virtual contact and friction force models

Abstract

In this paper, we compare many kinds of randomized algorithms and simultaneously inspect their data to calibrate an impulse force between two rigid bodies. One is an object handled by a manipulator tip, and another is an object fixed in a 3-D environment. A real impulse is obtained from an experiment via a six degrees-of-freedom force/moment sensor mounted on a tip of a seven degrees-of-freedom manipulator. As contrasted with this, a virtual impulse is quickly calculated in our impulse model. It consists of two classic models. One is Hertz's contact force model and another is Coulomb's friction force model. Each of extended randomized algorithms always minimizes the difference between real and virtual forces in order to get a better impulse model. In general, the randomized algorithm works well if and only if a set of correspondences between practical (true) and virtual (false) forces has high quality. Therefore, we carefully evaluate the correspondences and eliminate a few bad ones. Since our impulse model is not so complicated, it can be well calibrated from a few but good correspondences.