A virtual sensor approach to robot kinematic identification: theory and experimental implementation

Abstract

A system for identifying robot kinematics is presented which applies a particular virtual sensor approach to parameter estimation. The system incorporates the uncertainty of experimental measurements through the introduction of computationally simple variance matrix approximations. A line-fitting procedure which requires no additional sensors or accurately machined components is devised for experimentally comparing the parameter estimates with the robot design parameters. Experimental data were acquired from and the kinematic parameters are estimated for an Adept II robot manipulator using a commercially available visual 3D position sensor