Deep Reinforcement Learning based Path Planning for UAV-assisted Edge Computing Networks

Mobile edge computing (MEC) harvests the computation capability at the network edge to perform the computation intensive tasks for diverse IoT applications. Meanwhile, the unmanned aerial vehicle (UAV) has a great potential to flexibly enlarge the coverage, and enhance the network performance. Accordingly, it has been a promising paradigm to use the UAV to provide the edge computing service for massive IoT devices. This paper studies the path planning problem of a UAV-assisted edge computing network, where an UAV is deployed with an edge server to execute the computing tasks offloaded from multiple devices. We consider the mobility of devices, where a GaussMarkov random movement model is adopted. By taking the energy consumed for the dynamic flying and executing the tasks at the UAV into account, we formulate a path planning problem that aims to maximize the amount of offloaded data bits by the devices while minimizing the energy consumption of the UAV. To deal with the dynamic change of the complex environment, we apply the deep reinforcement learning (DRL) method to develop an online path planning algorithm based on double deep Q-learning network (DDQN). Extensive simulation results validate the effectiveness of the proposed DRL-based path planning algorithm in terms of the convergence speed and the system reward.