Path Following and Obstacle Avoidance for Autonomous Ground Vehicles Using Nonlinear Model Predictive Control

Abstract

This thesis presents a nonlinear model predictive controller (NMPC) for path following and obstacle avoidance in automated driving systems. Automated safety control systems have been increasingly effective at reducing the number of traffic fatalities in the United States. Many of the commercially available safety systems are still only classified as SAE level 1 and level 2 autonomy features. To progress towards SAE level 3 and level 4 automated driving systems, obstacle avoidance control must be added to the vehicle's dynamic driving task, removing human drivers from the control loop. Many current level 2 automated driving systems, such as Auburn University's heavy truck platooning system, could progress towards full autonomy by incorporating obstacle avoidance control into their existing control architectures. The NMPC control module developed in this work is designed to take advantage of an automated vehicle's existing software stack to provide enhanced path tracking and obstacle avoidance maneuvering.

Two simple vehicle models, a kinematic model and a dynamic bicycle model, are developed identified and implemented in a flexible NMPC software library which is feasible for real-time control of an autonomous vehicle. In a series of simulation and real-time experiments, a detailed tuning procedure and performance evaluation for both NMPC implementations are given. The kinematic model implementation is also shown to work as a replacement controller in Auburn University's existing software architecture for long distance, non-line-of-sight following of a manually driven leader vehicle. Obstacle avoidance is added to each controller implementation through a set of hard constraints. The feasibility of this constraint method is demonstrated with two simulated obstacle avoidance scenarios. Future improvements to both the obstacle avoidance method and path tracking accuracy are discussed.