Abstract
In the last decades, autonomous vehicles have been an active area of research in the academia as well as in industry. In particular, trajectory planning/control under emergency scenarios pose major challenges from a control point of view. In this thesis, we are addressing these challenges by developing controllers for trajectory planning/control for the application of autonomous collision avoidance system for the next generation of self-driving vehicles. The first part of this work is dedicated to design a trajectory planning/control under extreme driving conditions such as
high-speed driving and low surface friction conditions. It is imperative that the autonomous vehicles can ensure the safety considerations while performing a feasible and smooth manoeuvre. Thus we propose a framework including trajectory planing/control with integration of torque vectoring controller to design a feasible yet collision free trajectory subject to the external disturbances such as low surface friction and crosswind. Next, with the same control structure we propose a trajectory planing strategy using the idea of linear affine collision avoidance constraints that can be generalized for both straight and curve driving scenarios. A comparison between three controllers entitled as nominal model predictive control (MPC), offset-free MPC and robust MPC for planning a safe trajectory is then provided in order to justify the proposed trajectory planning design. Then the simulation results are validated in a high-fidelity co-simulation environment (IPG-Carmaker) to investigate the performance of the trajectory planning algorithms. In the second part of the thesis, a novel tuning technique is designed to improve the proposed control framework performance. The suggested tuning design is capable of providing optimal weights by matching the performance of a pre-design controller such as LQR controller to the trajectory planning/control framework. Finally, an intelligent controller entitled as Reinforcement Learning (RL) has been used as an on line tuning technique to improve the performance of yaw moment controller. This architecture is implemented on various operating conditions such as different friction surfaces and vehicle velocities and its performance has been validated using a four wheels vehicle model with non-linear tyre characteristic.