I. Introduction
State estimation is fundamental, notably for control purposes. This is particularly true for autonomous systems, such as autonomous cars. Prevailing approaches to the state estimation problem in robotics, and for inertial navigation in aerospace engineering, explicitly model sensor uncertainties, due to noise and bias, using the Gaussian random variables, and then seek to compute the maximum a posteriori (MAP) state, which is the most likely state in the light of all measurements while drawing on the vehicle’s dynamics. Such approaches allow combining sensor measurements optimally based on their confidence levels, as quantified by their covariance, and also allow the estimator to convey a degree of uncertainty associated with its own estimate, which may prove critical for high-level planning and low-level control of autonomous vehicles.