<b>Advances in Visual Simultaneous Localization and Mapping for Precise Lunar Landing</b>

<p dir="ltr">This thesis aims to improve visual navigation techniques for lunar landing. A single lidar scan of the lunar surface around a proposed landing site provides a starting point for a map, which is refined in-flight using robocentric square-root EKF-SLAM and square-root EIF-SLAM algorithms. The novel robocentric formulation provides an exact nonlinear relative motion framework between two moving objects, improving the consistency and observability of the filter SLAM approaches. Landmarks are selected for tracking according to a gradient-mapping technique based on the analytical expression of a weighted posterior covariance trace gradient. The robocentric SLAM algorithms with optimal landmark selection method are tested on a simulated lunar south pole descent and landing scenario, which shows excellent performance using only a handful of tracked landmarks. Alongside traditional consistency metrics, filter performance is evaluated with new whitened estimation error vector norm metrics, whose distributions (or approximations of) are derived for use in hypothesis testing. Following these results, avenues for future work are suggested based on the contributions of the thesis. </p>