A coding approach to localization using landmarks

Fully autonomous vehicles need the ability to localize without external help, for instance by using visual sensors together with a pre-loaded map of landmarks. In this paper we connect self-localization using landmarks with coding theory. This connection enables to translate Hamming distance properties to probabilistic localization guarantees given a certain number of errors in landmark identification; it also enables to leverage existing polynomial time decoding algorithms for localization. We present promising numerical evaluation results by simulating vehicle traveling paths along a road network generated from real data of a region in Washington D.C.