Navigating Complexity: A Deep Learning Approach to Lane Change Decisions in Autonomous Vehicles

Decision-making plays a critical role in guiding autonomous vehicles through complex environments characterized by dense traffic and dynamic elements. The existing research for decision-making in autonomous driving suffers from safety concerns, high computational power needs, low accuracy in comparison to human drivers, and their outputs cannot be easily interpreted. The paper discusses a deep learning-based decision-making algorithm for autonomous vehicles in dense and dynamic environments. It aims to create an easily interpreted algorithm that ensures safety, efficiency, and accuracy in lane change scenarios. The study introduces a novel approach that can be used within embedded systems and behaves similarly to human drivers. The methodology involves classifying lanes as blocked or free and making decisions based on this classification, mimicking human decision-making processes. The results show the algorithm’s effectiveness in complex scenarios, reaching a safety of 96% and accuracy of 90% which surpasses the Hierarchical Finite State Machine (HFSM) algorithm.