| The dynamic pose of an object, where the object can represent a spacecraft, aircraft, or mobile robot, among other possibilities, is defined to be the position, velocity, attitude, and angular velocity of the object. A new method to perform dynamic pose estimation is developed that leverages directional statistics and operates under the Bayesian estimation framework, as opposed to the minimum mean square error (MMSE) framework that conventional methods employ. No small attitude uncertainty assumption is necessary using this method, and, therefore, a more accurate estimate of the state can be obtained when the attitude uncertainty is large.;Two new state densities, termed the Gauss-Bingham and Bingham-Gauss mixture (BGM) densities, are developed that probabilistically represent a state vector comprised of an attitude quaternion and other Euclidean states on their natural manifold, the unit hypercylinder. When the Euclidean states consist of position, velocity, and angular velocity, the state vector represents the dynamic pose. An uncertainty propagation scheme is developed for a Gauss-Bingham-distributed state vector, and two demonstrations of this uncertainty propagation scheme are presented that show its applicability to quantify the uncertainty in dynamic pose, especially when the attitude uncertainty becomes large.;The BGM filter is developed, which is an approximate Bayesian filter in which the true temporal and measurement evolution of the BGM density, as quantified by the Chapman-Kolmogorov equation and Bayes' rule, are approximated by a BGM density. The parameters of the approximating BGM density are found via integral approximation on a component-wise basis, which is shown to be the Kullback-Leibler divergence optimal parameters of each component. The BGM filter is then applied to three simulations in order to compare its performance to a multiplicative Kalman filter and demonstrate its efficacy in estimating dynamic pose. The BGM filter is shown to be more statistically consistent than the multiplicative Kalman filter when the attitude uncertainty is large. |
