I. Introduction
Many robotic navigation applications require state-estimation methods beyond unimodal solutions due to four identified sources of non-Gaussian/multimodal behavior: i) non-linearity in measurement models [1]; ii) uncertain data association [2], [3]; iii) underdetermined problems (i.e. more unknowns than constraints) [4]; iv) physical measurement process is inherently ambiguous. This paper investigates latter two of these four identified sources, with the goal of developing robust simultaneous localization and mapping (SLAM). In addition, this paper addresses two major problems facing many perception systems, namely: handling of non-Gaussian data in an easy-to-understand factor graph framework, and methodologies that enable real-time navigation solutions. This work aims to demonstrate the feasibility, complexity, and importance of non-Gaussian solutions/methods in the context of applications dealing with such multimodal and highly uncertain problems.