A Comparison of Particle Filter and Graph-Based Optimization for Localization with Landmarks in Automated Vehicles

It is an essential capability of mobile robots and automated vehicles to localize themselves in a map. Multiple state estimation techniques exist to this end, and it is often unclear which one to employ. In this paper we compare two prominent techniques in the context of automated vehicles: particle filters and graph-based localization. The latter is the state-of-the-art approach to simultaneous localization and mapping, and is also superseding the particle filter as the state-of-the-art for pure localization. We compare both algorithms to show why. For both state estimation algorithms, we detect pole-like objects in laser scanner data and compare them against a prebuilt landmark map. The main novelty of this work is the experimental comparison on a real prototype vehicle. Furthermore, we discuss the different advantages of both algorithms in the context of automated driving and additionally show how both approaches can adapt their computational demand to the available resources at runtime.