AUTOMATIC PARAMETER IDENTIFICATION FOR GENERIC ROBOT MODELS

Abstract

High speed motion of robots and exact positioning demand of robots require highly accurate robot simulations. In existing generic models for different robot types, the choice of optimal parameters is a very important factor to obtain correct simulation results. The aim of this paper is to increase the accuracy of mechatronic simulations of generic robots by use of an automatic identification algorithm, which allows an easy identification of mechanical, drive and controller parameters. The use of algebraic least square methods based on dynamic equations is state of the art in robotics, however, different genetic algorithms have shown ex- cellent performance in many different applications in the past. In robotics the genetic algo- rithm is applied mainly in the area of trajectory optimization and the search of the optimal controller parameters. In the present paper a special automatic parameter identification al- gorithm, based on the principle of genetic optimization without parameter crossover, is de- scribed. Furthermore, a method is shown which considers multiple local minima of the simulation error. For verification of the algorithm the exactly known parameters of a simu- lated belt drive model are identified up to high accuracy. Finally, the algorithm is applied to measurement data of a real robot with parallel kinematics to identify certain drive parame- ters of the generic robot model, including the time delay of the measured torque. The simu- lated torque with optimized parameters shows high conformance with the real drive torque.