Unscented Filtering in a Unit Quaternion Space for Spacecraft Attitude Estimation

A novel approach to the straightforward implementation of unscented filtering in a unit quaternion space is proposed for spacecraft attitude estimation. A method of weighted mean computation for quaternions is derived in a rotational space, leading to a quaternion with a unit norm. Quaternion multiplication, which allows a quaternion in the filter to lie in the unit quaternion space, is then used for predicted covariance computations and quaternion updates. In this study, quaternion process noise, which increases the uncertainty in attitude orientation, is either modeled as the vector part of the quaternion or as a rotation vector. Simulation results indicate that the proposed approach successfully estimates spacecraft attitude.