Kalman filters for the identification of uncertainties in robotic contact

Abstract

The paper deals with the problem of identification of uncertainties in robotic contact operations. The uncertainties studied here are grasping uncertainties, contact uncertainties and friction. The principal tool used is the Kalman filter (KF). The stationary uncertainty vector is identified using an extended Kalman filter. The Kalman filter processes the force data from a wrist mounted force sensor. Depending on the contact situation, the KF sometimes fails to identify some of the uncertainties. Instead it can identify dependencies of the uncertainties. The paper describes how the estimation error covariance matrix of the KF can be used to identify the dependent degrees of freedom. Then it uses the identified dependent degrees of freedom to suggest how to change the contact situations so that all uncertainties are completely solved. Graphical results are presented showing how the KF progressed through different stages eventually solving for all the uncertainties.