An Online Dynamic Parameter Identification Approach for Robotic Manipulator With Reformulated Physical Feasibility

Accurate and stable identification of dynamic parameters is a critical challenge in the application of robotic manipulators, particularly for manipulators in dynamic environments. Existing methods often exhibit several problems: limited real-time adaptability, inadequate handling of physical feasibility constraints (PFCs), and insufficient robustness against uncertainties and external disturbances. To address these issues, this study presents a novel online dynamic identification method called recursive least-squares with PFC-guided update (RLS-PFC-G). The proposed approach integrates recursive least-squares (RLS), reformulated physical feasibility constraints (R-PFCs), and a nonlinear friction model. RLS enables real-time parameter updates, while the proposed R-PFC ensures positive definiteness of the inertia tensor and decouples it from the center of mass, thereby allowing for more precise and feasible parameter constraints. Additionally, a self-evolving fuzzy neural network (SE-FNN) is employed to mitigate uncertainties and external disturbances, bridging the gap between theoretical models and practical performance through torque compensation. Experimental results validated the effectiveness of RLS-PFC-G, demonstrating substantial improvements in identification accuracy and physical feasibility compared to conventional approaches.