Semantic-Aware Cooperative Edge Computing in UAV-Assisted Vehicular IoT

Abstract

Driven by the ongoing advancement of intelligent transportation systems (ITSs), conventional vehicular networks have rapidly evolved into a data-intensive vehicular Internet of Things (IoT) ecosystem. However, relying on traditional bit-level offloading in unmanned aerial vehicle (UAV)-assisted IoT networks creates severe physical channel bottlenecks when transmitting massive raw sensory data, significantly degrading system latency and energy efficiency. To address this challenge, we propose a digital twin (DT)-empowered, semantic-aware cooperative edge computing framework designed for highly dynamic vehicular IoT environments. In particular, we investigate a system cost minimization problem that jointly optimizes the semantic cache placement, task offloading, transmission power, and CPU frequency. To handle the non-Markovian dynamics and hybrid action space, we propose the federated semantic soft actor–critic (FedSSAC) algorithm, in which a long short-term memory (LSTM) network encodes the spatio–temporal state information, a soft actor–critic (SAC) architecture derives caching and resource allocation policies, and a federated averaging protocol enables privacy-preserving model aggregation. Simulation results demonstrate that the proposed framework achieves a high semantic cache hit rate and reduces both task processing latency and network energy consumption compared to conventional baselines.