Deep Reinforcement Learning Driven UAV-Assisted Edge Computing

Unmanned aerial vehicles (UAVs) are playing a critical role in provisioning instant connectivity and computational needs of Internet of Things Devices (IoTDs), especially in crisis and disaster management. In this work, we focus on optimizing trajectories of UAVs along which IoTDs are served with communication and computing resources in multiple time slots. The Quality of Experience (QoE) of an IoTD depends on its latency performance; we thus aim to maximize the average aggregate QoE of all IoTDs overall time slots. However, this is a nonconvex, nonlinear, and mixed discrete optimization problem, which is difficult to solve and obtain the optimal solution. We thus propose two deep reinforcement learning algorithms to solve this problem by considering UAV path planning, user assignment, bandwidth, and computing resource assignment. We compare the performance of our proposed algorithms through simulations with three baseline cases: 1) with fixed UAV locations; 2) without UAVs; and 3) the fixed UAV trajectories. We demonstrate that the deep reinforcement learning algorithms perform better than all baseline cases.