Smooth path planning for urban autonomous driving using OpenStreetMaps

Safe and human-like path planning in urban environments remain one of the most challenging problems for autonomous driving. This paper proposes a solution to plan traffic-free optimal and smooth paths using the most spread open source navigation data structure, OpenStreetMaps. To that end, a procedure to automatically transform raw OSM data into the drivable space is combined with a MINLP optimization algorithm that finds the most suitable position of intermediate waypoints to connect Bézier primitive curves. The proposed planner is validated in a real complex roundabout, for which very smooth paths are generated, even considering obstacles.