Contact-state Modeling of Robotic Assembly Tasks Using Gaussian Mixture Models

Abstract

This article addresses the Contact-State (CS) modeling problem for the force-controlled robotic peg-in-hole assembly tasks. The wrench (Cartesian forces and torques) and pose (Cartesian position and orientation) signals, of the manipulated object, are captured for different phases of the robotic assembly task. Those signals are utilized in building a CS model for each phase. Gaussian Mixture Models (GMM) is employed in building the likelihood of each signal and Expectation Maximization (EM) is used in finding the GMM parameters. Experiments are performed on a KUKA Lightweight Robot (LWR) doing camshaft caps assembly of an automotive powertrain. Comparisons are also performed with the available assembly modeling schemes, and the superiority of the EM-GMM scheme is shown with a reduced computational time.