Enhancing Dynamic Parameter Identification for Supernumerary Robotic Limbs Through Uncertainty Compensation

Accurate dynamic modeling is fundamental for wearable robotics, such as Supernumerary Robotic Limbs (SRLs), to ensure precise control and effective interaction with the environment. This paper presents an enhanced dynamic model for an SRL, incorporating uncertainty compensation to improve accuracy and robustness. The SRL dynamics are streamlined by linearizing the system and identifying a minimal set of inertia parameters. Using the Coot optimization algorithm, we refine the Fourier excitation trajectory parameters to optimize the condition number of the observation matrix, followed by parameter identification through the least squares method. To further reduce discrepancies between predicted and actual joint torques, an uncertainty term is introduced, validated through a neural network designed to predict torque variations. The method not only addresses but also compensates for uncertainties across various motion states of the SRL. Experimental results demonstrate that the proposed model improves prediction accuracy and robustness for our developed 6-DoF SRL compared to traditional methods.