Automated driving architectures using Model Predictive Control and Artificial Potential Fields

This thesis evaluates two automated driving approaches in several scenarios to compare their performance for standard driving maneuvers. Both approaches are based on the use of Artificial Potential Fields (APF) and Model Predictive Control (MPC), which computes the optimal control strategy by including the physical constraints of the vehicle. For one controller, the MPC computes directly the vehicle inputs in terms of steering angle and longitudinal acceleration. However, this approach is computationally demanding, primarily when implemented onboard the vehicle control units in a path planner, due to the intrinsic complexity of the underlying optimization problem. Thus, a second approach is considered, where a simplified MPC controller acts as a trajectory planner. A lower-level robust control strategy aims to follow the desired trajectory. Both these controllers are designed to reach the SAE automation level 3, i.e., an autonomous vehicle driving on a highway in everyday situations. Emergency maneuvers are left to the human driver and not managed from the control logic. Both implementations present advantages and disadvantages mainly based on their complexity, hence the importance of confronting both approaches under the same scenarios in this thesis. Thus, some simulated scenarios have been created to perform the expected maneuvers, i.e., lane keeping, distance tracking of the preceding vehicle, overtaking, and re-entry. The parameters of both kinds of controllers have been tuned to maximize the performances and offer a fair comparison between the two approaches.