Two Stage Identification of Robotic Dynamic Parameters with the LuGre Friction Model using Improved Grey Wolf Optimizer Method

Abstract

In order to develop a manipulator dynamic model to predict the torque behavior, a two-stage approach is proposed in this paper. First, the excitation trajectory is used to identify the link parameters of the robotic manipulator, such as the mass, length and inertial, and Coulomb friction parameters through least squares (LS), a generic algorithm (GA), and the Improved Grey Wolf Optimizer (I-GWO) method. Then the dynamic components computed from the identified parameters are removed from the torque commands. Subsequently, the friction described by a nonlinear LuGre friction model is proposed for the second stage and identified subsequently. Experimental results demonstrate that the approach could predict the torque command more accurately with the LuGre model. Moreover, the maximum error of the torque command during joint velocity changes in motion direction could be reduced by $42.17 \%$. The averaged error for the testing trajectory could improve torque command accuracy by about $15.19 \%$.