DT-MASAC: A Digital-Twin-Augmented Multiagent Framework for AoI-Oriented Resource Allocation in Vehicular Networks

Vehicular edge computing (VEC) networks supporting delay-sensitive services face fundamental challenges in maintaining information freshness under high mobility, bursty task arrivals, and limited communication and computation resources. This paper investigates an Age of Information (AoI)-oriented resource allocation problem in digital twin (DT)-empowered VEC systems, where task offloading, wireless transmission, and computation scheduling are tightly coupled across time. To address the limitations of reactive and myopic optimization, we propose a DT-augmented multi-agent soft actor–critic (DT-MASAC) framework that explicitly embeds AoI dynamics and DT-based predictive features into cooperative multi-agent learning. The DT layer provides short-term foresight of channel conditions and edge resource availability, enabling agents to make proactive and uncertainty-aware decisions. A joint AoI–energy minimization problem is formulated under realistic communication, computation, and deadline constraints, and solved via centralized training with decentralized execution. Extensive simulations demonstrate that the proposed DT-MASAC framework achieves superior AoI stability and energy efficiency compared with state-of-the-art baselines, particularly in highly dynamic vehicular environments.