Autonomous Parking Using Previous Paths

Abstract

This paper is about mapping the drivable area of a parking lot for autonomous parking. Manual map creation for automated parking is often impossible, especially when parking on private grounds. One aspect is that the number of private properties is very large and private parking should not be included in public maps. The other aspect is that an owner and operator of a car often has very specific ideas of where the car may be driven. Our approach creates maps using just the previously driven paths. We describe the drivable area through triangles using established methods from Computer Graphics. These triangles are generated by overlaying circles of a certain radius over the driven paths. These circles create a so-called alpha-shape and approximate the drivable area. The description through triangles (“Delaunay triangulation”) allows for fast retrieval and easy expansion with new paths. Finally, a simple conversion of the triangulation into a Voronoi diagram enables fast path searching. In this paper we thus present an efficient framework for determining drivable areas and allows searching for a drivable path. Finally, we show that this method enables real-time implementation in an autonomous car and can cope with new obstacles at planning time.