Automated Valet Parking in eco-system contexts: an ego-based approach via Model Predictive Control

In the context of automated driving, Automated Valet Parking (AVP) is a functionality that aims at parking and retrieving vehicles in a suitably equipped infrastructure. In recent years, AVP showed its great potential in optimizing parking spaces, fuel consumption and reducing driving stress and time waste for parking maneuvers. From a methodological standpoint, the main contribution of this thesis is the development of an ego-based AVP approach. In such a context, the infrastructure provides a minimum amount of information about the parking structure. On the other hand, the needed intelligence is allocated to the vehicle, which handles path planning and tracking, environmental awareness, obstacle avoidance and parking maneuvers. A precedence system was also devised, allowing to handle complex, multi-vehicle scenarios at road intersections. The perception and sensing problem is developed under the assumption that all the needed information about the surrounding environment is obtained from the vehicle's on-board vision system and sensors. An AVP functionality must generate and track collision-free trajectories to drive the vehicle from the drop-off area to the parking site and back to the pick-up zone. These tasks are accomplished by an architecture that includes three interconnected modules, i.e. the Global Planner, the Local Planner and the Parking Planner, coordinated by a Behavioral Logic on the basis of sensor data. The Global Planner computes an optimal geometric path from the drop-off area to one of the assigned bays, combining Dijkstra's algorithm and Dubins' curves. This module includes a "Global-in-the-loop" unit that dynamically evaluates the possibility of changing path or destination whenever an obstacle is detected on the trajectory. The Local Planner tracks the path generated by the Global Planner through a Nonlinear Model Predictive Controller (NMPC) based on Artificial Potential Fields (APFs). Such a controller handles obstacle avoidance, includes an Adaptive Cruise Control (ACC) mode and imposes driving physical and comfort constraints. The Parking Planner generates a feasible parking maneuver to be tracked by a NMPC controller from the proximity position to the parking bay. The Behavioral Logic interacts with the other modules and is in charge of analysing sensor data to perform obstacle detection, obstacle identification and decision making. It also oversees the precedence system and decides when to invoke the global-in-the-loop. The main advantage of the adopted strategy is that, it is designed to operate with a minimal amount of information provided by the infrastructure, but any additional V2X communication can be incorporated to enhance its performance. Since all the intelligence is allocated to the vehicle, the AVP functionality is compliant with infrastructures spanning a wide range of technological facilities. Furthermore, as the different modules are organized as an encapsulated architecture, the approach offers great adaptability for integration purposes within the control systems of different vehicles. To show the effectiveness of the proposed approach, extensive and realistic simulation tests in different complex, multi-vehicle scenarios are performed considering a real parking architecture and a nonlinear vehicle model.