Particle Filtering for Attitude Estimation Using a Minimal Local-Error Representation

Spacecraft attitude estimation is the process of determining the orientation of the spacecraft with respect to a reference frame. Many nonlinear filtering methods have been applied to spacecraft attitude estimation. The extended Kalman filter (EKF) has been the most widely used nonlinear filtering method for realtime attitude estimation so far [1]. It works well in the linear regime where the linear approximation of the nonlinear dynamical system and observation model is valid. An unscented Kalman filter (UKF) has been proposed for attitude estimation, which uses a set of deterministically chosen sigma points to more accurately map a probability distribution and which is more robust than the EKF under large initial attitude-error conditions [2]. Other improved nonlinear filtering algorithms for attitude estimation are surveyed in [3]. Most of these algorithms estimate the mean and covariance of the state vector based on secondor higher-order approximations of nonlinear functions. Although the mean and covariance are the sufficient statistics of a Gaussian distribution, they are not sufficient to represent a general probability distribution. When these methods are applied to strongly nonlinear and non-Gaussian estimation problems, where the probability distribution of the state vector may be multi-modal, heavy-tailed, or skewed, desired performance characteristics may not be obtained. Particle filters (PFs) [4] have gained much attention in recent years. Like other approximate nonlinear filtering methods, the ultimate objective of the PF is to reconstruct Assistant Professor, Department of Aerospace Engineering. Email: [email protected]. Senior Member AIAA. Professor, Department of Mechanical & Aerospace Engineering. Email: [email protected]. Associate Fellow AIAA.