An adaptive UKF algorithm and its application for satellite attitude determination

The normal unscented Kalman Filter(UKF) suffers from performance degradation and even divergence while mismatch between the noise distribution assumed to be known by UKF and the true ones in a real systems. Based on maximum a posterior(MAP), a modified noise statistic estimator was proposed, keeping the estimated noise covariance positive define matrices within some rules. The adaptive filtering is realized by changing and computing the noise statistics on line. Under the condition of unknown noise statistic, this method is able to compensate the errors by updating mean and covariance of the noise, retraining the filter's divergence, moreover its filtering precision is better than conventional UKF. Simulations conducted on the satellite attitude determination system indicate that the adaptive UKF is superior to the conventional UKF in terms of estimation accuracy and stability.