Adaptive Unscented Kalman Filter for online state, parameter, and process covariance estimation

Abstract

A novel observer for state, parameter and process covariance estimation is presented in this paper. The new observer estimates system states using a Square-Root Unscented Kalman Filter (SRUKF) and by employing the Recursive Prediction-Error (RPE) method, unknown parameters and covariances are identified online. Two experimental applications based on an underactuated planar robot are included to demonstrate the algorithm performance. Additionally, sensitivity models for the SRUKF are derived. Results show that the online process covariance estimation improves the observer convergence and reduces parameter estimation bias.