A Separable Instrumental Variable Method for Robot Identification

Abstract

This paper addresses the topic of robot model identification. The standard identification method is based on the use of the inverse dynamic model (IDM) and the application of Least Squares (LS) estimation while the robot is tracking trajectories. Although this approach has been applied successfully to several industrial robots, the standard friction model is assumed to be described by a linear relationship in the viscous and Coulomb coefficients and the consistency of LS estimates is not secured when the system is identified in closed loop. In this paper, a Separable Instrumental Variable (SIV) approach is proposed that corrects these limitations. This SIV method consists of two-steps. In the first one, a Nonlinear LS estimation is combined with the IDM to identify the nature of friction effect, which is potentially nonlinear. In a second step, an iterative linear IV method is performed in order to obtain the estimates of the inertia and gravity parameters. The experimental results obtained on the 6 degrees-of-freedom TX40 robot show the effectiveness of this new approach.