Edge Computing Frameworks for Real-Time Optimisation in Autonomous Electric Vehicle Networks

Abstract

Autonomous electric vehicles (AEVs) require real-time decision-making, low-latency computation, and energy-aware coordination to operate effectively. Traditional centralised cloud computing struggles to meet these demands due to inherent delays and scalability issues in large-scale AEV networks. This paper proposes a novel hybrid edge–fog computing architecture to address these challenges. Our framework utilises a three-tier system (vehicle-edge, roadside-fog, and cloud) governed by a deep reinforcement learning agent that manages energy-aware task offloading. Extensive simulations demonstrate the framework's effectiveness, achieving significant end-to-end latency reductions of up to 56% during urban peak hours and decreasing energy consumption by 20% under high-load conditions. The deep reinforcement learning agent successfully adapts control policies to dynamic road conditions, while the architecture proves highly scalable and resilient, maintaining high task success rates and recovering from node failures in seconds. These findings confirm that a hybrid edge–fog architecture, guided by reinforcement learning, is a highly effective solution for scalable, adaptive, and energy-efficient AEV operations. This study's primary contribution is an empirically validated framework that uniquely integrates predictive control and energy-aware scheduling at the edge, providing a deployable model for next-generation intelligent transportation systems.