Design and Implementation of Parking Control Algorithm for Autonomous Valet Parking

This paper represents a parking lot occupancy detection and parking control algorithm for the autonomous valet parking system. The parking lot occupancy detection algorithm determine the occupancy of the parking space, using LiDAR sensors mounted at each side of front bumper. Euclidean minimum spanning tree (EMST) method is used to cluster that information. After that, a global parking map, which includes all parking lots and access road, is constructed offline to figure out which cluster is located in a parking space. By doing this, searching for available parking lots has been finished. The proposed parking control algorithm consists of a reference path generation, a path tracking controller, and a parking process controller. At first, route points of the reference path are determined under the consideration of the minimum turning radius and minimum safety margin with near parking. After route points is determined, the reference path is generated by connecting straight lines and arcs between pre-determined route points. A path tracking controller which determines a desired steering wheel angle is designed by the combination of road curvature based feedforward and feedback linear quadratic optimal control method using 2 DOF bicycle model. In this paper, a desired velocity of vehicle is maintained as constant during the parking process to simplify the path generation problem. Finally, parking process control algorithm determines the sequence of parking and gear change timing. The proposed control algorithm for autonomous valet parking has been validated via computer simulations and successfully implemented on a test vehicle.