Deep Reinforcement Learning-Based Autonomous Navigation Systems for Intelligent Transportation Networks

Abstract

The rapid advancement of autonomous vehicles and intelligent transportation networks (ITNs) has transformed modern mobility systems, promising improved safety, efficiency, and sustainability. Deep reinforcement learning (DRL), an intersection of deep learning and reinforcement learning, has emerged as a powerful computational paradigm for enabling autonomous navigation, traffic optimization, and adaptive decision-making in dynamic urban environments. This research investigates DRL-based autonomous navigation systems, focusing on their integration into intelligent transportation networks to enhance vehicle routing, collision avoidance, and traffic flow management. A hybrid methodology combining simulation-based experiments and computational modeling was employed to evaluate the performance of DRL agents in multi-agent traffic scenarios. Structural equation modeling using SmartPLS was applied to examine the relationships between algorithm design parameters, environmental perception accuracy, decision-making efficiency, and navigation outcomes. Results indicate that DRL algorithms significantly enhance autonomous navigation performance (β=0.74, p<0.001) by optimizing route selection, reducing collisions, and improving travel time reliability. Environmental perception accuracy (β=0.68, p<0.001) and decision-making efficiency (β=0.71, p<0.001) mediate the relationship between DRL algorithm sophistication and navigation outcomes, emphasizing the importance of sensor fusion, real-time data processing, and adaptive reward mechanisms. The study demonstrates that DRL-based navigation systems provide robust, scalable, and adaptive solutions for autonomous vehicles operating within complex ITNs, outperforming traditional rule-based and classical path-planning methods. Implications include enhanced traffic efficiency, reduced congestion, and increased safety for autonomous transportation networks. Future research should explore integration with vehicle-to-everything (V2X) communication, edge computing, and multi-agent reinforcement learning to further optimize real-time traffic coordination. The findings provide both theoretical and practical insights for researchers, transportation planners, and policymakers seeking to deploy DRL-driven autonomous navigation systems for next-generation intelligent transportation networks.