UAV-aided wireless networks for vehicular communications

In recent years, Unmanned Aerial Vehicles (UAVs) have emerged as versatile platforms with potential across a wide range of domains, from environmental monitoring and emergency response to logistics and communications. Among their most promising applications is their integration into wireless networks as aerial communication nodes, leading to the development of truly three-dimensional (3D) architectures. The use of UAVs equipped with base-station capabilities, namely Unmanned Aerial Base Stations (UABSs), offers a flexible and rapidly deployable complement to traditional terrestrial infrastructures. Operating at elevated altitudes, UABSs can overcome propagation obstacles, ensuring line-of-sight (LoS) connections, and delivering on-demand capacity in critical areas. These characteristics make them particularly attractive for future 6G vehicular communication systems, where Vehicle-to-Everything (V2X) applications will demand high levels of reliability, latency, and continuity of service that terrestrial networks alone may not sustain. This thesis investigates the design of autonomous and scalable trajectory-planning algorithms for UABSs operating in coordination with terrestrial networks to support vehicular communications. The objective is to enable intelligent and energy-efficient aerial operations capable of providing continuous and adaptive connectivity in highly dynamic environments. The research first establishes a foundation for joint trajectory and radio resource management (RRM) through a deep reinforcement learning (DRL) framework allowing a UABS to adapt its flight path in real time based on vehicular mobility. The study then introduces a modular DRL architecture that incorporates dynamic beamforming activation, enabling the UABS to balance energy consumption and communication performance within constrained energy mission budgets. Then, a cooperative multi-UABS framework based on multi-agent DRL (MADRL) is developed, featuring a proactive quality-of-service-aware anti-collision mechanism that ensures safe and coordinated flight in dense deployments. Finally, the thesis explores multi-scenario adaptability through meta-learning, allowing trained agents to rapidly adapt their trajectories to new traffic conditions and service maps without full retraining, thus enabling efficient and flexible deployment.