Asymptotically Optimal Attitude Filtering with Guaranteed Convergence

Abstract

This paper introduces some new concepts in attitude filtering to aid the formulation of simple filtering algorithms with guaranteed convergence from almost any initial condition, far outperforming the transient performance of standard extended Kalman filter formulations. The focus is on formulating a filter algorithm for estimating the four quaternion elements using unit vector observables to provide periodic updates of gyroscope-propagated attitude. It is assumed that the gyroscope rate measurement errors are solely due to zero-mean noise. This algorithm is suitable for coarse attitude determination using elementary sensors. Research into extending the approach to more complicated error models is ongoing.