A Comparative Study of Deep Reinforcement Learning Applications for Autonomous Train Operation

This paper explores the application of Deep Reinforcement Learning (DRL) for optimizing Automatic Train Operation (ATO). The study aims to enhance train operations by meeting key objectives such as energy efficiency, punctuality, safety, passenger comfort, and parking accuracy. A simulation environment incorporating realistic train dynamics and track conditions was developed, utilizing discrete action and observation spaces. Various DRL algorithms, including QRDQN and PPO, were implemented and tuned using Optuna, a hyperparameter optimization framework. Results demonstrate QRDQN's superior performance across multiple evaluation journeys, highlighting its potential for achieving robust, efficient, and autonomous train operation.