A constrained instrumental variable method for identification of industrial robots

Abstract

Robot identification is a prolific topic with a long history of results spanning decades. In recent years, there has been a renewal of interest in this problem mainly due to a rapid increase in robotic hardware platforms capable of accurate model-based control. The standard approach exploits the inverse dynamic model’s linearity to dynamic parameters and uses the linear least-squares (LS) estimation. Since we identify robots with closed-loop procedures, a correlation between errors remains, and we should prefer the Instrumental Variable (IV) method over the LS estimation. Thanks to the increase in computational power, recent works suggest inserting physical constraints to ensure the physical plausibility of estimates. These works have emphasized the usefulness of these physical constraints, but few papers consider their insertion into IV methods, the consistency and optimality of estimates, and the effect of constraints on estimates not addressed. This paper presents a new constrained IV approach that uses physical constraints. It consists of two nested iterative algorithms: an outer one that is a standard IV method and an inner one that accounts for the constraints solved by a Gauss–Newton algorithm. Besides, the conditions to obtain consistent and optimal estimates are emphasized. Experimental results and comparisons with other methods carried out with the TX40 robot show the feasibility and effectiveness of such an IV method: we can identify 60 physically consistent parameters in less than one minute.