Section: New Results

Evidential Simultaneous Localization And Mapping to describe intersection

Participants : Guillaume Trehard, Evangeline Pollard, Fawzi Nashashibi.

Intersections management remains a tough challenge to tackle before reaching autonomous driving in urban environment. The field of view of the vehicle is often limited by several sensors occlusions, the shapes and priority rules can significantly differ from an intersection to another and road users from pedestrian to public transports have to cross each other in sometimes complex manners.

In this context, mapping the surrounding of the vehicle and being able to estimate its position regarding a global database is crucial.

A solution of Simultaneous Localization And Mapping (SLAM) have then been proposed based on a 2D LIDAR sensor [49] . In the rich SLAM literature, the originality of this method lays in the use of Transferable Belief Model (TBM) framework instead of a classic probabilistic one. If this proposition was just a change of mathematical context, TBM led to an explicit management of not-known and conflict information so that its application to SLAM algorithm appeared to be really effective and robust in crowded situations. The proposed solution indeed enables to provide a map of the static environment crossed by the vehicle and to detect mobile obstacles in the same process and without additional tracking system.

This Evidential SLAM have then been tested with success over different sequences and laser set-ups extracted from the KITTI database [50] .

Researches are now focused on the fusion between this SLAM solution and a Global Navigation Satellite System (GNSS) receiver to enable a map-matching on a database such as Open Street Map.