Ego-based Automated Valet Parking Control: a Model Predictive Control approach

In the context of autonomous driving, Automated Valet Parking (AVP) is a function that parks and retrieves vehicles in a suitable parking infrastructure. In recent years, AVP became an interesting topic for researchers due to its several advantages such as optimization of parking spaces, fuel consumption, reduction of driving stress and time saving for parking maneuvers. The main contribution of this thesis is the development of a vehicle ego-based approach, i.e., the infrastructure provides a minimum amount of information about the parking structure, while the necessary intelligence for automated valet parking is allocated to the vehicle, which takes care of path planning and tracking, obstacle avoidance and parking maneuver. The problem of AVP is developed with the assumption that all the needed information about the surrounding environment is available from the on-board vision system and sensors. To realize the AVP function, a collision-free trajectory is generated to drive the vehicle from the drop-off area to the parking site. Such a trajectory is obtained through a three stages procedure that includes the development of a Global Planner, a Local Planner and a Parking Planner. The Global Planner computes a minimum distance geometric path from the drop-off area to the parking proximity position by combining Dijkstra’s algorithm and Dubins’ Curve. The Local Planner tracks the path generated by the Global Planner through a Nonlinear Model Predictive Controller (NMPC), based on Artificial Potential Fields (APFs), avoiding obstacles and imposing driving physical and comfort constraints. The Parking Planner, generates a feasible geometric path to drive the vehicle from the parking proximity position to the parking site. Such a path is tracked by a NMPC controller. A similar procedure is employed to drive the vehicle from the parking site to the pick-up area. The main advantage of the strategy introduced in this thesis is that, since all the intelligence is located into the vehicle, the automated parking procedure is infrastructure independent and the vehicle can exploit AVP function in every parking structure. Extensive simulation tests are realized considering a nonlinear vehicle model and different possible parking scenarios to show the effectiveness of the proposed approach.