Age of Information Optimization in Laser-charged UAV-assisted IoT Networks: A Multi-agent Deep Reinforcement Learning Method

TL;DR

This paper proposes a multi-agent deep reinforcement learning framework to optimize the age of information in laser-charged UAV-assisted IoT networks, balancing data freshness and energy efficiency.

Contribution

It introduces MAPPO-TM, a novel multi-agent DRL method with temporal memory for AoI optimization in UAV IoT networks with laser charging.

Findings

Achieves up to 15.1% reduction in peak AoI.

Outperforms conventional MADRL methods.

Enhances energy efficiency and coordination among UAVs.

Abstract

The integration of unmanned aerial vehicles (UAVs) with Internet of Things (IoT) networks offers promising solutions for efficient data collection. However, the limited energy capacity of UAVs remains a significant challenge. In this case, laser beam directors (LBDs) have emerged as an effective technology for wireless charging of UAVs during operation, thereby enabling sustained data collection without frequent returns to charging stations (CSs). In this work, we investigate the age of information (AoI) optimization in LBD-powered UAV-assisted IoT networks, where multiple UAVs collect data from distributed IoTs while being recharged by laser beams. We formulate a joint optimization problem that aims to minimize the peak AoI while determining optimal UAV trajectories and laser charging strategies. This problem is particularly challenging due to its non-convex nature, complex temporal dependencies, and the need to balance data collection efficiency with energy consumption constraints. To address these challenges, we propose a novel multi-agent proximal policy optimization with temporal memory and multi-agent coordination (MAPPO-TM) framework. Specifically, MAPPO-TM incorporates temporal memory mechanisms to capture the dynamic nature of UAV operations and facilitates effective coordination among multiple UAVs through decentralized learning while considering global system objectives. Simulation results demonstrate that the proposed MAPPO-TM algorithm outperforms conventional approaches in terms of peak AoI minimization and energy efficiency. Ideally, the proposed algorithm achieves up to 15.1% reduction in peak AoI compared to conventional multi-agent deep reinforcement learning (MADRL) methods.