Comparative Study of Model-based Lateral Controllers with Selected Deep Learning Methods for Autonomous Driving

Road accidents, predominantly caused by human errors like distracted or drunk driving, are significant threats to transportation safety. Autonomous vehicles, using classical approaches like the Pure Pursuit Controller, Stanley Controller or MPC offer a potent solution. These traditional methods, though effective, depend on intricate vehicle dynamics equations, making them prone to inaccuracies. On the other hand, with the advent of Artificial Intelligence, End-to-End Deep Learning techniques like PilotNet, and its more advanced versions employing residual blocks and LSTMs, present a promising alternative too. However, there is little literature comparing these two Lateral Control approaches against each other and with manual human driving. This thesis seeks to fill that gap by contrasting these models' efficacy in predicting steering angles under consistent road conditions and trajectories, thereby providing insight on which technique/s are optimal for real-time autonomous vehicle navigation.