AHSWN Home · Issue Contents · Forthcoming Papers

Multi-Agent Deep Learning for QoS-aware Task Scheduling in UAV-assisted Wireless Ad Hoc Networks
Heming Fan, Ying Xu, Qian Liu, Yapeng Gao, Zi Wang, Amin Mohajer and Lifen Gu

Unmanned Aerial Vehicles (UAVs) have become instrumental in expanding network coverage and ensuring connectivity in wireless ad hoc networks. However, the dynamic nature of these networks, characterized by changing topologies, limited resources, and diverse traffic, presents significant challenges. A critical aspect of managing these networks is ensuring Quality of Service (QoS) while optimizing task scheduling to achieve efficient resource utilization and reliable communication. To address these challenges, we propose a novel approach, Multi-Agent Deep Learning for QoS-aware Task Scheduling (MADL-QTS). Our framework utilizes multi-agent deep reinforcement learning (MADRL) to dynamically allocate resources and make informed task scheduling decisions. MADL-QTS offers several key contributions to the field: (1) Decentralized resource allocation: by modeling the network as a multi-agent system, each UAV functions as an autonomous agent, learning optimal policies for task scheduling. This decentralized approach enhances scalability and adaptability to ever-changing network conditions. (2) QoS-driven policies: our approach prioritizes QoS satisfaction, with the QADRA scheduler optimizing for real-time QoS metrics such as latency, reliability, and throughput. This focus ensures that critical applications receive the necessary attention, leading to improved network performance. (3) Context-aware scheduling: UAV agents are equipped to adapt to user mobility and specific application requirements, making dynamic adjustments based on real-time conditions to improve responsiveness and resource utilization. By addressing the trade-offs between energy efficiency and low latency, MADL-QTS revolutionizes task scheduling in UAV-assisted wireless ad hoc networks. Empowering UAV agents with deep reinforcement learning, context-awareness, and QoS-driven policies, our approach significantly enhances network efficiency and user experience.

Keywords: UAV-assisted wireless ad hoc networks, multi-agent deep learning, QoS-aware task scheduling, decentralized resource allocation, context-aware scheduling