Particle Filtering for Sequential Spacecraft Attitude Estimation

Abstract

A new spacecraft attitude estimation approach using particle filtering is derived. Based on sequential Monte Carlo simulation, the particle filter approximately represents the probability distribution of the state vector with random samples. The filter formulation is based on the star camera measurements using a gyro-based or attitude dynamics-based model for attitude propagation. Modified Rodrigues parameters are used for attitude parametrization when the sample mean and covariance of the attitude are computed. The ambiguity problem associated with the modified Rodrigues parameters in the mean and covariance computation is addressed as well. By using the uniform attitude probability distribution as the initial attitude distribution and using a gradually decreasing measurement variance in the computation of the importance weights, the particle filter based attitude estimator possesses global convergence properties. Simulation results indicate that the particular particle filter, known as bootstrap filter, with as many as 2000 particles is able to converge from arbitrary initial attitude error and initial gyro bias errors as large as 4500 degrees per hour per axis.