Path tracking control solutions via Enhanced Model Reference Adaptive Control algorithms augmented with Neural Networks and their experimental validation in scaled fully autonomous vehicles

Autonomous driving represents a pivotal and extensively researched technology, characterized by features that are progressively finding their way into commercial vehicles. This technology relies on a combination of sensors, actuators, and sophisticated software, seeking to improve the safety and reliability of vehicles, aiming at the replacement of human drivers. This work targets the path tracking problem i.e. the automatic steering of the vehicle in order to follow a specific path with no human action; this task is undertaken in both simulated and real experimental environments. The employed control algorithm is the Enhanced Model Reference Adaptive Control (EMRAC), it is an adaptive control design method that allows the controlled variables of a plant to track a given reference model. In this project, two types of EMRAC control systems were devloped to guide a real, scaled, and fully autonomous vehicle prototype along a predetermined path: a standard EMRAC and an EMRAC enhanced with a Neural Network. The controllers have been designed for a wide range of velocities showing notable improvements with respect to benchmark controllers as well as the benefits of the Neural Network based augmentation. This research project contributes to reducing the gap in experimentally validated Enhanced Model Reference Adaptive Control algorithms present in the existing literature.