A MADRL-Based Trajectory Planning Scheme for Multi-UAV Assisted VLC Systems with Mobile Users

Unmanned aerial vehicles (UAVs) offer significant advantages, including rapid deployment and extensive coverage capabilities. These benefits have drawn considerable attention to their application in the field of communications, particularly in scenarios such as the internet of things (IoT), emergency communications, and disaster relief. Additionally, visible light communication (VLC) is gaining interest due to its high data transmission rates and ability to mitigate spectrum scarcity issues. Integrating VLC with UAVs combines these strengths, creating UAV-assisted VLC systems that provide robust and efficient communication solutions in various challenging environments. However, most of the existing research on UAV-assisted VLC only consider scenarios with static users, instead of mobile users. In this study, we propose a novel trajectory planning scheme based on multi-agent deep reinforcement learning (MADRL), specifically, using the algorithm of multi-agent proximal policy optimization (MAPPO), which enables UAVs to dynamically adjust their flight paths in real-time to maintain stable and efficient VLC links with mobile users. Our approach not only improves the efficiency of UAV path planning but also enhances system adaptability to dynamic environments, offering robust support for optical communication. This scheme significantly boosts the communication capabilities of UAVs while reducing their energy consumption, addressing a critical gap in current research which primarily focuses on static or simple environments. Simulation experiments validate the proposed MADRL-based scheme, and demonstrate that MAPPO outperforms other baseline algorithms, consistently delivering superior performance across all metrics.