Markov Localization Using Correlation

Abstract

konolige @ ai.sri.com Localization is one of the most important capa-bilities for autonomous mobile agents. Markov Localization (ML), applied to dense range im-ages, has proven to be an effective technique. But its computational and storage requirements put a large burden on robot systems, and make it difficult to update the map dynamically. In this paper we introduce a new technique, based on correlation of a sensor scan with the map, that is several orders of magnitude more efficient than M L. CBML (correlation-based ML) permits video-rate localization using dense range scans, dynamic map updates, and a more precise error model than M L. In this paper we present the ba-sic method of CBML, and validate its efficiency and correctness in a series of experiments on an implemented mobile robot base. 1