Constructing contextual SLAM priors using architectural drawings

Abstract

Accurate robot mapping, localisation and navigation remains an unsolved problem for challenging real-life indoor environments. Many approaches to Simultaneous Localisation And Mapping (SLAM) have been proposed but few attempts have been made to improve performance by using appropriate prior maps. Information such as floor plans or architectural drawings is available and there is a rich literature of processing floor plans to extract information. However, the problem of converting drawings to an appropriate SLAM prior format has not been addressed. This paper addresses this problem and proposes a way to process such plans using a simple set of geometric constraints to extract useful information and construct appropriate SLAM priors. It also proposes a set of criteria and a method to assess the quality of constructed SLAM priors.