Priority-Aware Hybrid Actor-Critic for Task Scheduling of UAV-Assisted Edge Computing

Abstract

Unmanned Aerial Vehicle (UAV)-assisted Mobile Edge Computing (MEC) reduces network latency by leveraging the high mobility and flexibility of UAVs to provide on-demand computing services for heterogeneous user terminals (UTs). However, scheduling tasks across UAVs is challenging due to the diverse priority requirements of UTs, ranging from ultra-low-latency control tasks to high-throughput video streaming. To address this, we propose PAHAC, a Priority-Aware Hybrid Actor-Critic algorithm with a continuous-discrete action encoder. PAHAC embeds discrete scheduling decisions into a differentiable action space for joint optimization with continuous control variables, while incorporating priority weights and deadline penalties into the reward function. The maximum entropy framework, dual critics, and adaptive entropy coefficients further enhance exploration and convergence stability. Simulation results demonstrate that PAHAC achieves lower latency, higher throughput for highpriority tasks, and greater scheduling stability across low-to-high load conditions, confirming its effectiveness for priority-aware task scheduling in UAV-assisted MEC systems.