Study of predictive control for trajectory following in an autonomous vehicle

Abstract

This thesis investigates the design, implementation, and comparative evaluation of advanced predictive control techniques for the trajectory tracking on the AutoNOMOS-Model-v3.1 autonomous vehicle in the GNC project framework. Its primary aim is to enhance path-following accuracy and robustness by integrating a nonlinear Model Predictive Control (NMPC) algorithm and a Linear Quadratic Regulator (LQR) within a ROS-based framework, which the GNC project is based on. The first chapter focuses on explaining the theoretical background needed for the project. This includes a brief review of MPC and LQR, including the deduction of LQR’s implicit solutions, the description of the kinematic, including its linearization, and the dynamic model used to describe a vehicle’s motion, along with a brief description of splines, focusing particularly on those types that are commonly used for path planning. The second chapter explains the necessary ROS concepts needed to understand the rest of the thesis along with a brief description of the ROS and code structure of the GNC project. Next, there is a detailed explanation of the software developed during the project. This includes the development of the two controllers, simulation environment, and data storage, along with the modifications made in the path determination algorithm. Simulation results are very promising for the MPC controller developed, although the LQR and MPC results in the autonomous car show that more work needs to be done to mitigate sensor inaccuracies and odometry drift, which degrade the performance. Overall, the study achieves its initial objectives and provides a guide and foundation for further development.