Celestial navigation is a commonly used autonomous navigation technique for deep space navigation. A nonlinear filter such as the unscented Kalman filter (UKF) is typically applied in a celestial navigation system (CNS). However, on account of being subject to a number of factors, such as ephemeris errors and centroid determination, the measurement model error of a CNS cannot be accurately determined. The analysis conducted in this study also shows that the measurement model error is time-variant during the Mars approach phase. This implies that covariance matrix of the measurement error R is usually inaccurate, which may induce large estimation errors that even result in filter divergence. Some adaptive methods are able to address this issue. However, traditional adaptive filters, for scaling R, usually require a sequence of innovation and are affected by the statistic window size. A new recursive adaptive UKF (RAUKF) is proposed in this paper, which only uses current innovation to scale R. The navigation performance of the proposed RAUKF method is compared with some traditional adaptive filters through simulations. The results show that this method is better than traditional adaptive filters in a CNS during the Mars approach phase.
