Force Feedback Control in Insertion Process using Pattern Analysis Techniques

Abstract

A statistical analysis of peg-hole configuration recognition during the peg insertion process is presented. Equations for generating a force/moment vector corresponding to each possible contact configuration are derived. These equations are used to generate training samples for learning the distribution parameters of the conditional probabilty density function for each peg-hole configuration. A decision rule, based on minimum probability of the error, is then formulated for recognition of the peg-hole contact configurations. Results of a computer simulation of the proposed learning and recognition techniques are presented. Experimental verification of the technique is currently being conducted on a PUMA robot arm.