Artificial Potential Fields-based Predictive Control for Autonomous Vehicles in Highway Scenarios

Autonomous driving represents nowadays a great challenge and goal for both the industry and the academic world due to its numerous advantages such as traffic optimization, improved safety, lower emissions and passengers comfort. The key to accomplish this goal is a synergic work between the on-board vision system and sensors, responsible for the interaction with the driving scenario, and the control system, that implements the autonomous tasks. This thesis focuses on the problem of autonomous local navigation within highway scenarios, by assuming that the necessary information coming from on-board sensors is available. Such a problem is divided into two main steps: finding a optimal path and then performing its tracking through a suitable control system. As a first step, a higher level controller that employs Artificial Potential Fields(APFs) methods for path planning is developed and casted in a Nonlinear Model PredictiveControl (NMPC) framework. In this context, a deep study on the specific properties of the APFs needed to accomplish each task, i.e., vehicle following, overtaking, is performed to account for all the travelling performances required in highway driving, in terms of safety and comfort. After the definition of suitable time-varying potential fields virtually superimposed on the driving environment, a NMPC controller generates the trajectory to be tracked by the vehicle. In the second step, the trajectory computed in the upper level is tracked by a lower level controller. Such a low level control logic is made up by two decoupled contributions: a PID controller to handle the longitudinal dynamics, and a LQR controller to regulate the lateral dynamics. Extensive simulation tests performed on a nonlinear vehicle model are introduced to show the effectiveness of the proposed approach in several highway driving scenarios. Future works include both the real-time implementation of the proposed path planning control architecture, and actual on-field experiments to validate the performances obtained in the simulation tests.