A new nonlinear filtering method and its application to attitude estimation

Predictive filter is a new filter which can be used to nonlinear systems with large model error. In this filter, there is an important parameter- weighting matrix of model error, which is difficult to compute but greatly affects the final estimation results. To address this problem, an improved predictive filter was presented. The superiority of the new filter is providing an efficient method to compute the weighting by using a self-regulate scheme. This scheme is implemented in real-time by changing the weighting matrix with the covariance of the model error. Theoretically, the reasonability of the modified filter was analysed. As application, the new method was applied to attitude estimation. Its effectiveness and superiority over the original filter was demonstrated by simulation.