Multi-Agent Graphic Reinforcement Learning for Real-Time UAV Video Transmission with Predictive Target Tracking

In the area of real-time video transmission, multi-hop unmanned aerial vehicle (UAV) networks offer promising solutions, especially in places where traditional equipment cannot be deployed. However, the mobility of UAVs introduces a large amount of link uncertainty, which is often ignored by existing methods. To address this issue, this paper introduces a dynamic predictive system that can adapt to dynamic conditions and provide high-quality transmission. Additionally, to address the critical challenge of limited onboard energy in video transmission, we propose a two-scale strategy focusing on resource allocation and target tracking. Specifically, the strategy divides the prediction of dynamic target trajectory trends and resource allocation into two scales and considers them respectively. To achieve great prediction and two-scale strategies, we propose the graph multi-agent proximal policy optimization (GM-PPO) algorithm. This algorithm combines the strengths of graph-based learning with multi-agent systems, ensuring efficient resource utilization, enhancing communication reliability, and optimizing trajectory planning. Simulation results demonstrate that our approach can efficiently solve the joint problem of power allocation and trajectory optimization while satisfying the delay constraint.