Adaptive Unscented Kalman Filter at the presence of non-additive measurement noise

This paper proposes an Adaptive Unscented Kalman Filter (AUKF) for nonlinear systems having non-additive measurement noise with unknown noise statistics. The proposed filter algorithm is able to estimate the nonlinear states along with the unknown measurement noise covariance (R) online with guaranteed positive definiteness. By this formulation of adaptive sigma point filter for non-additive measurement noise, the need of approximating non-additive noise as additive one (as is done in many cases) may be waived. The effectiveness of the proposed algorithm has been demonstrated by simulation studies on a nonlinear two dimensional bearing-only tracking (BOT) problem with non-additive measurement noise. Estimation performance of the proposed filter algorithm has been compared with (i) non adaptive UKF, (ii) an AUKF with additive measurement noise approximation and (iii) an Adaptive Divided Difference Filter (ADDF) applicable for non-additive noise. It has been found from 10000 Monte Carlo runs that the proposed AUKF algorithm provides (i) enhanced estimation performance in terms of RMS errors (RMSE) and convergence speed, (ii) almost 3-7 times less failure rate when prior measurement noise covariance is not accurate and (iii) relatively more robust performance with respect to the initial choice of R when compared with the other nonlinear filters involved herein.