Most of Current autonomous navigation solutions critically rely on SLAM systems for localisation, especially in GPS-denied environments but also in urban and indoor environments. Their efficacy and efficiency thus depend on the ability of the underlying SLAM method to map large-scale environments in a data-efficient manner. State of the art systems, while accurate, often require powerful hardware such as GPUs, and need careful tuning of hyperparameters in order to adapt to the user's needs. In this paper, we propose a lightweight but accurate probabilistic 2D graph-based SLAM system. We validate our approach on sequences from the KITTI dataset as well as on data gathered by our experimental platform.