Age-Driven Joint Optimization for UAV Swarms via Multi-Agent Reinforcement Learning

Real-time monitoring in remote areas presents critical challenges for time-sensitive missions such as emergency response and disaster surveillance, necessitating autonomous UAV swarms with stringent information timeliness requirements. For this end, we adopt a Leader-Follower UAV swarm architecture, in which Follower UAVs act as both sensing platforms and communication relays, thereby establishing a fully airborne network without reliance on ground infrastructure. To enhance information timeliness, we adopt an age-driven formulation quantified by the Age of Information (AoI), within which sampling decisions, buffer scheduling, and routing selection are jointly optimized under dynamic topology and partial observability. Accordingly, we model this problem as a Decentralized Partially Observable Markov Decision Process (Dec-POMDP), and develop an enhanced multi-agent reinforcement learning (MARL) algorithm that employs specialized policy networks to decompose coupled decision-making into coordinated sub-tasks. Simulation results demonstrate the proposed approach significantly reduces average AoI compared to baseline algorithms while maintaining competitive performance in conventional network metrics including packet delivery ratio (PDR) and throughput.