Robust Parameter Estimation of Robot Manipulators Using Torque Separation Technique

Abstract

In this paper, we propose a robust method for estimating the parameters of robot manipulators using the torque separation technique, which was developed previously by the authors, to extract the inertial, gravitational, and frictional components from the input torques of multiple sinusoidal joint motions. The separated components of the input torque produce a set of reduced linear regression equations where the dynamic parameters of the robot manipulators appear decoupled or minimally coupled. A mathematical analysis is presented to show that the set of reduced regression equations tends to yield more robust parameter estimation than the conventional way where a large dimensional full regression equation is solved simultaneously. An iterative scheme to alleviate the possibility of parameter estimation error caused by the deviation from the assumed ideal sinusoidal motions is also proposed. Experimentation with two robot manipulators is used to verify the proposed approach and related claims. The proposed method is simple and pragmatic without requiring any specialized signal filtering and technical know-hows which numerous previous methods often demanded for explicitly or implicitly.