Uncorrelated unscented filtering for spacecraft attitude determination

Spacecraft attitude estimation based on the nonlinear unscented filter is addressed to fully utilize capabilities of the unscented transformation. To overcome significant computational load, an efficient technique is proposed by appropriately eliminating correlation between random variables. This modification leads to a considerable reduction of computational burden in matrix square-root calculation for most nonlinear systems. The unscented filter makes use of a set of sample points to predict mean and covariance. For attitude estimation based on quaternions, an approach to computing quaternion means from sampled quaternions with guarantee of the normalization constraint is described by using a constrained optimization technique. Finally, the performance of the new approach is demonstrated by attitude determination using a star tracker and rate-gyro measurements.