Hierarchical Model Predictive Control for Trajectory Generation and Tracking in Highly Automated Vehicles

Abstract

The research field of autonomous vehicle technology has been growing at an accelerated pace. Improved safety, fuel and commuting efficiency are the motivating technical and social factors to develop fully autonomous vehicles. Sensor technology, advanced software and intelligent control are the different modules that work in unison to achieve the desired driving result. The system architecture of autonomous vehicles is explored to establish a hierarchy with the planning and control stages. This allows us to focus on the particular topic of intelligent control for both levels of operation. In this thesis, a hierarchical model predictive control is developed for trajectory generation and tracking of on-road vehicles. The state of the art methods in planning and control are predominantly developed in the robotics domain and the additional challenges of the vehicle such as non-linear dynamics, sampling time and limited computational resources make it a challenging problem. $10$ DOF, six state model, and point mass vehicle dynamics models was evaluated to sufficiently represent the dynamics of the systems and allow for efficient operation in the controller development phase. Model predictive control (MPC) is chosen because of its capability of systematically taking into account non-linearities, future predictions, and operating constraints during the control design stage. In the hierarchical approach, at the high-level, new trajectories are computed on-line, in a receding horizon fashion, based on a simplified point-mass vehicle model in order to avoid the obstacle. The formulation of the collision avoidance constraints to render a quadratic programming (QP) problem from a non-convex optimization problem was crucial in the trajectory generation phase. The parameter values involved in the forward and read collision avoidance constraints defined the feasible driving regions. At the low-level an MPC controller computes the vehicle control inputs steering and acceleration in order to best follow the high level trajectory based on a higher fidelity non-linear vehicle model. The simulation scenarios defined cases for static obstacle avoidance, car following and special overtake manoeuvres. The effectiveness of the controllers were strongly affected by the parameter tuning of the vehicle, design constraints, and collision avoidance terms. The chosen method implemented a hierarchical controller with a higher level deliberative paradigm and lower level tracking controller to achieve the tasks with respect to highly automated driving.