Deep Reinforcement Learning for Efficient Data Collection in UAV-Aided Internet of Things

Abstract

In the Internet of Things (IoTs), unmanned aerial vehicle (UAV) has been considered as an efficient solution to collect information from ground sensor nodes (SNs) due to its controllable mobility and high maneuverability. In this paper, we study a UAV-aided efficient data collection problem for IoTs, where SNs sample information with fixed or random rates and cache the sampled update packets under a sample-and-replace policy. An energy-constrained UAV is deployed to collect data from each SN when it flies over this SN. The UAV's flight trajectory is optimized to minimize the SNs' average age-ofinformation (AoI) while preserving their packet drop rate as low as possible. Toward this end, we model the UAV-aided data collection problem as a finite-horizon Markov decision process (MDP) with finite state and action spaces. Then, we develop a deep reinforcement learning algorithm to find an asymptotically optimal policy. Simulation results demonstrate that the proposed learning algorithm can significantly reduce the AoI and packet drop rate, compared to the baseline greedy algorithms.