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

A review of particle filters (PF) and quaternion attitude kinematics and gyro and vector observation models is provided. An attitude estimation PF using a local-error representation is provided and PF is compared with the extended Kalman filter (EKF) and unscented Kalman filter (UKF) using simulated three-axis magnetometer (TAM) and gyro measurements of an Earth-pointing spacecraft. It is found that the conversion from the local-error representation to the global representation requires less computations than implementing a strategy that explicitly maintains quaternion normalization. The particle filter is based on the bootstrap filter, still the local/global representation can be applied to any particle filter formulation. Simulation results indicated that the performance of the new filter based on the idea of progressive correction exceeds the standard extended Kalman and unscented Kalman filter for large initialization errors.