SORTING PARTS BY RANDOM GRASPING

As a low-cost alternative to machine vision, we consider how a modifie d parallel-jaw gripper can be used to classify parts according to shap e by grasping and measuring the diameter: the distance between the jaw s. Since more than one part may give rise to the same diameter and the sensor may be corrupted by noise due to surface compliance and backla sh, we show how the most probable part can be estimated using a sequen ce of random grasps with a Bayesian decision procedure. This procedure allows us to define a statistical measure of the ''similarity'' of a set of parts. Laboratory experiments confirm that the random strategy is effective for sorting parts.