Real-time isolation of physiological tremor using recursive singular spectrum analysis and random vector functional link for surgical robotics.

Abstract

Hand-held robotic instruments enhance precision in microsurgery by mitigating physiological tremor in real time. Current tremor filtering algorithms in these instruments often employ nonlinear phase prefilters to isolate the tremor signal. However, these filters introduce phase distortion in the filtered tremor, compromising accuracy. Although improved variants of recursive singular spectrum analysis (RSSA) have addressed the issue of phase distortion, they still face challenges such as reduced generalization performance, large sample delays, and longer computational times. To address these issues, we integrate an accurate and fast random vector functional link (RVFL) with RSSA, referred to as RSSA-RVFL. The proposed approach consists of two main steps: estimation using RSSA and prediction with RVFL. Additionally, we introduce two moving window variants of RSSA-RVFL for real-time implementation. These variants significantly reduce computational costs while delivering the same performance. Experimental results on real tremor data show that our proposed approach achieves an average accuracy of 79.03%, surpassing the benchmark of 70.40%, with a nine-sample delay.