Ola Rønning will gives a talk on a filtering approach to simultaneous localization and mapping.
SPEAKER: Ola Rønning, PostDoc, ITU.
TITLE: Real-Time Bayesian Filtering for SLAM with Stein Mixtures
ABSTRACT: Robots navigating unstructured environments must estimate their pose and map their surroundings in real time while accounting for uncertainty. Classical methods such as Kalman and particle filters capture uncertainty but rely on restrictive assumptions that limit accuracy and scalability. We propose a Bayesian filtering approach based on Stein mixtures, aiming to balance accuracy and uncertainty while scaling to the full SLAM state. Although still in development, we present preliminary results on synthetic SLAM scenarios that demonstrate improved accuracy over Kalman filtering, and we analyze the convergence rate needed for Stein mixtures to be competitive with classical approaches.