Development and Preliminary Evaluation of a RANSAC Algorithm for Dynamical Model Identification in the Presence of Unmodeled Dynamics

Abstract

This paper reports a novel Random Sample Consensus (RANSAC) algorithm for robust identification of second-order plant dynamical model parameters in the presence of unmodeled plant dynamics and noisy experimental data. Accurate plant dynamical models are essential to model-based control system design and for accurate numerical simulation of plant response. Studies of RANSAC approaches for plant model identification have been extremely limited and have not explored performance improvements in the presence of unmodeled dynamics. The performance of the proposed approach, evaluated in a preliminary simulation study of a planar aerial rotorcraft model, is found to be significantly more robust to the effects of unmodeled vehicle dynamics and outlier noise than conventional least squares parameter identification. We conjecture that the proposed approach may be broadly applicable to robust model parameter identification for a wide variety of plants that exhibit noisy sensor data and/or unmodeled dynamics.