Random execution of a set of contacts to solve the grasping and contact uncertainties in robotic tasks

Abstract

This paper addresses the problem of identifying the uncertainties present in a robotic contact situation. These uncertainties are errors and misalignments of an object with respect to its ideal position. The paper describes how to solve for the errors caused during grasping and errors present when coming into contact with the environment. A force sensor is used together with Kalman Filters to solve for all the uncertainties. The straightforward use of a force sensor and the Kalman Filters is found to be effective in finding only some of the uncertainties in robotic contact. The other uncertainties form dependencies that cannot be estimated in this manner. This dependency brings about the problem of observability. To make the unobservable uncertainties observable a sequence of contacts are used. The error covariance matrix of the Kalman Filter (KF) is used to obtain new contacts that are required to solve for all the uncertainties completely. There is complete freedom in choosing which unobservable quantity to be excited in forming the next contact. The paper describes how these new contacts can be randomly executed. A two dimensional contact situation will be used to demonstrate the effectiveness of the method. Experimental data are also presented to prove the validity of the procedure. Due to the non-linear relationship between the uncertainties and the forces, an Extended Kalman Filter (EKF) has been used.