LAUSR

Recently, unmanned aerial vehicle (UAV) technology is endorsed to enable applications in domains such as Internet of things (IoT), wireless sensor networks, and cellular networks. Particularly, time-sensitive and energy-limited IoT networks located in hard-to-reach areas require efficient/cost-effective data collection solution. To address this matter, we consider a multi-UAV enabled IoT network, where several UAVs collect data from time-constrained sensor nodes (SNs). In our framework, SNs are managed by cluster heads (CHs), then UAVs collect data from them. We formulate the problem of minimizing system’s deployment costs and operating energy to collect data within deadlines, subject to communication, UAVs mission time, and battery capacity constraints. To solve it, we propose a two-step approach. In the first step, an efficient K-means based method groups SNs and deploys CHs. Then, UAV-based offline and online data collection methods are proposed. In the offline setting where the system’s status is known beforehand, UAV paths are determined using near-optimal meta-heuristics. In simulations, the nearest-neighbor and Tabu search provided best offline performances, conditioned on the system’s parameters. In the online setting where no system information is available, deep reinforcement learning (DRL) based approaches are proposed. Results demonstrate the superiority of the actor-critic solution.