Parameter identification of a robot arm using separable least squares technique

This paper presents the application of the separable least squares technique to the parameter estimation of a six-degrees-of-freedom robot arm. A dynamic model of the robot that is linear in parameters is obtained, and a joint friction model including both linear and nonlinear terms is adopted. Linear least squares methods can not be used here because of the nonlinear friction terms. Once the excitation trajectories - which strongly influence the quality of estimation - are optimized, the robot is excited in closed loop and the collected experimental data are used to estimate all inertial and friction parameters using the separable least squares technique. The main contribution of this paper is to propose a time-efficient method to estimate the linear and nonlinear parameters of robot arms simultaneously. The obtained model is validated in two experimental tests: a torque prediction and a trajectory tracking task using a model-based inverse dynamics controller. The results of both tests performed on the CRS A465 robot arm demonstrate the high accuracy of the estimated model. Moreover, it is shown that including the Stribeck friction term has clearly improved the model accuracy.