Explainable Deep Reinforcement Learning for Autonomous Transportation Systems

Abstract

Deep reinforcement learning has emerged as a powerful computational framework for autonomous transporta tion systems where vehicles learn to navigate, coordinate, and make decisions through continuous interaction with dynamic road environments. Despite its effectiveness, the opaque nature of learned policies poses significant challenges for reliability, safety assurance, and operational transparency. This article presents a comprehensive study of explainable deep reinforcement learning applied to autonomous transportation tasks. A multi layer architecture is introduced that integrates interpretable state attribution, policy visualization, and reward decomposition techniques into the learning pipeline. The framework was evaluated using simulated mobility scenarios with varying road layouts, congestion levels, and interaction patterns. The results show that explainability mechanisms improve decision traceability while sustaining competitive performance across navigation, collision avoidance, and cooperative driving tasks.