Autonomous Parking Maneuvers via Non-Linear Model Predictive Control

Autonomous driving is considered one of the most ground-breaking technologies of the near future that will completely reshape transportation systems. In this regard, more and more research efforts are being spent by automotive companies and academic institutions for developing vehicles with an ever-higher level of autonomy. Thanks to well-known environments and relatively low risks, particular attention has been devoted to Autonomous Parking. The aim is to simplify as much as possible the actions required by the driver to complete the parking, reducing time (to perform all the maneuver) and spaces. Indeed, the increase in the number of vehicles entails the need for even narrower spots, thus rendering manual parking operations more challenging. Furthermore, unskilled parking abilities may cause traffic jams. Modern control theory offers a multitude of approaches and design paradigms that can be exploited for this application. Among them, Nonlinear Model Predictive Control (NMPC) has the potential to become a key technology. Certainly, it: i) only requires a target point, ii) deals with linear and nonlinear constraints, iii) jointly performs trajectory planning and control. In that context, this thesis aims to develop a general NMPC framework capable of performing several parking maneuvers. More in detail, a suitable configuration of the parameters characterizing the NMPC has been found such that, providing different initial poses and targets, it is always able to generate the optimal trajectory and commands to guide the ego vehicle into the parking zone. Furthermore, to avoid collisions, Heaviside-based constraints have been used. Finally, to prove the effectiveness and robustness of the developed NMPC framework, a Monte Carlo campaign has been carried out considering different initial conditions. In all the performed tests, the NMPC succeeded in entering the parking zone without any collision.