PML-SLAM: a solution for localization in large-scale urban environments

Localization is considered a key factor for autonomous cars. In this paper, we present a complete Simultaneous Localization And Mapping (SLAM) solution. This algorithm is based on probabilistic maximum likelihood framework using grid maps (the map is simply presented as a grid of occupancy probabilities). The solution mainly solve three renowned localization problems (1. localization in unknown environment, 2. localization in a pre-mapped environment and 3. recovering the localization of the vehicle). Memory issues caused by the open size of outdoor environment are solved using an optimized management strategy that we propose. This strategy allows us to navigate smoothly while saving and loading probabilities-grid submaps into/from a hard-disc in a transparent way. We present the results of our solution using our own experimental dataset as well as the KITTI dataset.