Dynamic Identification of Industrial Robots from Low-Sampled Data

Abstract

This paper proposes a fast and on-site method for the dynamic identification of industrial robots from low-sampled position and torque data. Owing to the basic architecture of the employed controller, only trapezoidal-velocity trajectories can be enforced for identification purposes. Differently from previous literature, where this kind of trajectories were performed with limited joint velocities and range of motions, the procedure proposed hereafter is characterized by fast movements performed on wide angular ranges. Furthermore, in order to identify the influence of friction without deriving complex friction models, a novel method is outlined that decouples frictional torques from gravitational, centrifugal and inertial ones. Finally, although multiple experiments of different kinds have been performed, inertial parameters are determined in one singular step, thus avoiding possible error increase due to sequential identification algorithms.