The attitude control problem, or the control of a spacecraft's orientation with respect to a frame of reference, is a challenging problem in space missions and has attracted much attention as it involves highly nonlinear characteristics of the governing equations. The attitude control task requires an estimation algorithm that deduces the attitude from strapdown sensor inputs and a control algorithm that computes the necessary torques so that the vehicle can follow a desired attitude.

From the perspective of control, feedback control laws are sought for the purpose of asymptotic trajectory tracking, with the ability to reject unexpected external disturbances, and be insensitive to parameter variations. An adaptive sliding mode spacecraft attitude controller that fulfills those requirements is discussed in this dissertation. Unit quaternions and Rodrigues parameters are used to parameterize attitude. Lyapunov stability theory is used to prove the stability of the closed-loop system.

For attitude estimation with increased accuracy, strap-down gyroscopes and vector measurements are fused together. Because of the nonlinear nature of the attitude kinematics equation and the measurement model, the problem becomes a nonlinear state estimation problem, which is typically tackled by Bayesian inference. In this dissertation we discuss a marginalized particle filtering algorithm, to possibly increase the estimation accuracy and reduce the computation load compared with other non-parametric methods. We exploit the linear-substructure and further show that the linear state evolution is completely independent of the nonlinear partition.

We have also investigated a computationally efficient and easy-to-tune sensor fusion algorithm, based on the complementary filter and the TRIAD algorithm. It is beneficial to use a complementary filter because rate and angle sensor possess benefits and drawbacks in different frequency regimes. The proposed algorithm shows comparable performance to the EKF but with less computational burden. It aims to be implementable on a small portable platform. In applications of mobile robots, the cutoff frequency can be adapted based on a fuzzy logic in real-time to adjust trust to different sensors, to cope with problems such as motion accelerations and magnetic distortions.