Improved Dynamic Model Identification of Cobots Based on a Nonlinear Friction Model

In the field of advanced manufacturing with human robot collaboration (HRC), accurate dynamic model of cobots plays a crucial role in model-based control, simulation, and safe physical human-robot interaction (pHRI). However, excessive noise and inaccurate friction modeling often lead to unsatisfactory identification results. To address these challenges, this paper improves dynamic model identification by adopting a nonlinear friction model that captures both the Stribeck effect and the nonlinear characteristics of viscous friction, offering a more precise representation of frictional behavior across various velocity ranges. Furthermore, an iterative identification scheme is employed to to obtain the optimal physically feasible parameters of the dynamic model, thereby enhancing the model's robustness and performance. Building on the identified physics-based model, we introduce a clear learning-based method that integrates physical knowledge into the black-box model to enhance friction modeling. Through experimental analysis of six different methods, it is demonstrated that the proposed method outperforms the others.