Contact localisation: a novel approach to intelligent robotic assembly

Abstract

The successful mating of mechanical components can only be achieved by determining appropriate corrective actions to compensate for initial angular and linear misalignment. The automation of assembly processes calls for systems which provide a flexible and nonlinear behaviour due to the large uncertainties intrinsic in the selection of a correct action. These issues indicate strongly that an artificial neural network can lead to effective controllers in the field of robotic assembly. In this paper we describe the design of an intelligent assembly controller based on the fuzzyART neural network. This unsupervised classifier provides fast and stable behaviour that proves capable of merging geometrical and sensorial data into the estimation of the contact location. This in turn is used to enable a planned list of actions to be followed during the insertion of a prismatic peg in a matching hole. The real time implementation of the proposed architecture has identified the issues related with its practical applicability.