Existing dynamic model identification methods are difficult to simultaneously address both the inertial parameters and friction of the model subject to physical feasibility. This paper proposes a hybrid physics-data driven dynamic modeling framework (H-PDD) that resolves the inherent coupling between LuGre friction characteristics and inertial parameter identification. It synergizes physical models with data-driven approaches to perform parameter optimization, ensuring physically consistent solutions while enhancing identification precision. Firstly, the Iteratively Reweighted Least Squares (IRLS) algorithm initializes the inertial part of the base parameter set. Then, this framework constructs distributed friction observers using physics-informed neural networks (PINN) to explicitly embed LuGre friction micro-deformation states. By formulating torque balance equations, it implicitly couples inertial torque components with nonlinear friction forces. Finally, a regularization constraint based on the positive definiteness of the inertia matrix is designed to optimize inertia and friction parameters synchronously for global optimal matching. For a 6-DOF robot, the H-PDD algorithm is compared with seven other traditional algorithms on two test trajectories. The results show that the proposed algorithm has higher accuracy in terms of root mean square error and R-squared.
Xu et al. (Fri,) studied this question.