Application of Reinforcement Learning Models for Optimization of Air Traffic Control Protocols

This research explores how SARSA, Expected SARSA, and Q Learning algorithms can automate vital decisions in air traffic control (ATC) during aircraft en-route. These models are equipped with Reinforcement Learning (RL) techniques that enable them to adapt to changes in airspace conditions as well as sequence aircrafts by considering parameters such as heading, speed and altitude Different scenarios were used for extensive testing and Q-Learning demonstrated superior performance over SARSA and Expected SARSA with a margin of approximately 14.89% and 9.65% respectively. Additionally, Q-Learning was found to be capable of sequencing an average of fifteen aircrafts per episode.