Path-Planning for Autonomous Parking with Dubins Curves

Autonomous cars can offer numerous services to its users, one is fully autonomous parking. There are different challenges involved in autonomous parking on parking spaces such as efficient path planning under constraints or limited sensing of today’s cars. In this paper, we present a path planning approach for automated car parking in unstructured environments. Our system is able to find paths without imposing additional restrictions on either the environment, the final parking position, the number of direction switches, nor the length of the path. Our method consists of a lattice grid search, which yields kinematically-feasible paths and a subsequent optimization step to obtain a dynamically-desirable solution. The main contribution of this paper is in the construction of edges for the grid search through Dubins Curves. We implemented and throughly tested our system both in simulation and on a real Mercedes-Benz E-Class instrumented for automated driving. The parking solutions are found quickly and allow for an accurate execution so that parking is possible for smalland medium-sized parking lots. We believe that this paper is a viable approach towards fully automated parking.