Dynamic Parameter Identification for Redundant Manipulator Using a Hierarchical Framework With Nonlinear Compensation

A high-precision dynamic model is crucial for enhancing the control performance and improving the motion accuracy of robots. However, traditional algorithms can only integrate a linear friction model, leading to limited identification accuracy and yielding dynamic parameters that are not physically feasible or lack rationality. To address these problems, this study proposed a four-step hierarchical identification framework. Initially, all dynamic parameters were identified by incorporating physical feasibility constraints and Riemannian geometry metrics. A hierarchical method was then developed to integrate and reidentify various nonlinear friction parameters. Subsequently, the inertial parameter reidentification refined the parameter distribution, which enhanced the overall accuracy. To compensate for uncertain identification errors, an extensive dataset was used to train a Gaussian process regression model, thereby further enhancing the identification accuracy. This methodology addressed three key limitations—that is, the incompatibility with nonlinear friction models, low identification accuracy, and failure to satisfy physical feasibility constraints. Extensive comparative experiments were conducted using eight distinct algorithms. Compared to the classical weighted least squares algorithm, the identification and validation accuracies improved by an average of 16.84% and 22.94%, respectively. The results indicate that the proposed method exhibited more reasonable parameter values, a more accurate friction model, and improved identification accuracy.