Optimizing the Average Age of Information in Multi-UAV-Assisted MEC Systems with Stochastic Task Arrivals

Unmanned aerial vehicle (UAV)-assisted mobile edge computing (MEC) has gained significant attention as an effective solution in real-time mission-critical applications such as emergency response and disaster relief operations to enhance computational capabilities and reduce service latency. This paper proposes a novel multi-UAV cooperative MEC system that addresses the challenges posed by stochastic task arrivals from ground users, while enabling collaborative computing offloading among UAVs. The primary objective is to minimize the overall average age of information (AAoI) for all users under joint constraints on UAV trajectory optimization and computation offloading allocation. Our key innovation lies in the development of a federated learning framework with parameter-sharing mechanisms, where UAVs and ground users operate as heterogeneous multi-agents with in a distributed training architecture. By implementing the multi-agent actor-critic (MAAC) algorithm, the proposed approach enables online optimization of systemwide AAoI through collaborative policy learning. Comprehensive experimental evaluations demonstrate that the proposed scheme achieves lower AAoI and greater stability compared to conventional MAAC (without parameter sharing) and multi-agent deep deterministic policy gradient (MADDPG) approaches, particularly in dynamic environments with high task arrival rates.