6G THz-Semantic Access for Federated Learning Enabled Internet of Unmanned Vehicle Agents

With the advent of 6G and the evolution of vehicular communication networks, the Internet of Vehicles (IoV) is facing unprecedented demands for ultra-high data rates and ultra-low latency to support advanced autonomous driving applications. However, deploying terahertz (THz) communication in such networks presents significant challenges. First, the massive exchange of raw sensor data with semantic information among vehicles creates severe communication bottlenecks, thereby undermining the real-time decision-making required by autonomous systems. Second, inherent characteristics of the THz band—such as high attenuation and pronounced molecular absorption—substantially restrict the effective transmission range, particularly in shortrange real-time scenarios, thus imposing stringent requirements on the timeliness of driving decisions. To address these challenges, we propose a Semantic THz Communication Empowered Wireless Federated Learning framework for unmanned vehicle decision-making. Our approach leverages semantic compression to extract and transmit only task-relevant information, thereby dramatically reducing the data volume and alleviating network congestion. Moreover, by integrating wireless federated learning into the IoV architecture, vehicles collaboratively train robust autonomous driving models while preserving data privacy. This distributed learning mechanism—by treating each vehicle as a node that contributes to the overall model—effectively compensates for the short-range limitations of THz links by harnessing their ultra-wideband capability. Numerical results demonstrate that our proposed framework significantly improves throughput and reduces latency compared to conventional mmWave-based approaches, ultimately enhancing the accuracy and efficiency of driving decisions.