Computation Offloading Service in UAV-Assisted Mobile Edge Computing: A Soft Actor-Critic Approach

This paper investigates a system where an unmanned aerial vehicle (UAV) provides computing services to multiple mobile terminals. To optimize the computation offloading services of the system, we jointly optimize mobile terminal scheduling, UAV trajectory, and task offloading ratio to minimize service delay and terminals' energy consumption under energy constraints. Considering the optimization problem with non-convexity, the high-dimensional states, and continuous action space, we propose a SAC-based computation offloading algorithm (SAC-CO) for UAV-assisted mobile edge computing (MEC). The algorithm improves its convergence by combining off-policy and maximum entropy reinforcement learning. Simulation results demonstrate that SAC-UTO outperforms the benchmark algorithm in most cases.