Trajectory Design in UAV-Aided Mobile Crowdsensing: A Deep Reinforcement Learning Approach

Abstract

Mobile crowdsensing (MCS) is a method of data collection by recruiting mobile devices to accomplish various sensing tasks. The mobility and intelligence of mobile devices enable an efficient solution to large-scale sensing, e.g., smart city. Unmanned aerial vehicles (UAVs), as mobile devices, can be used in MCS to perform many sensing tasks (e.g., monitoring). In addition, UAVs provide new business opportunities (e.g., package delivery) with its rapid increasing number. We aim to leverage the package delivery activities of UAVs to solve the task allocation problem of MCS. In the package delivery activities, UAVs must deliver the assigned packages to their destinations. During the package delivery, UAVs travel around to perform sensing tasks with time windows. In this case, the task allocation problem of MCS is considered as a trajectory design problem of UAVs. To plan the trajectories of UAVs, we propose a deep reinforcement learning approach, specifically, double deep Q-network with prioritized experience replay (DDQN-PER). Finally, the results of our numerical simulations show that our proposed solution outperforms two baseline solutions in terms of profit and number of completed tasks.