Graph-based SLAM using wall detection and floor plan constraints without loop closure

This paper describes a graph-based SLAM approach using wall detection and floor plan constraints without relying on loop closure. In SLAM, loop closure is widely used to address cumulative errors. Although loop closure helps maintain the map’s relative consistency, it does not ensure the accuracy of absolute positions. Therefore, we focus on floor plans that accurately depict the environmental geometry and propose a SLAM method that leverages this information. However, floor plans do not depict semi-static objects such as bookshelves and other fixtures. Thus, our study aims to build accurate maps based on floor plans and represent actual environments. The proposed method achieves this goal by integrating wall detection and floor plan constraints within the framework of graph-based SLAM. We evaluated the proposed method based on qualitative assessments of mapping results and quantitative evaluations of robot trajectories and processing time. Experiments were conducted using datasets obtained from both simulation and real-world environments. The results demonstrate that the proposed method can build a map with accurate absolute positions in a low processing time by leveraging wall detection and floor plan constraints.