AoI-Aware Data Collection in Heterogeneous UAV-Assisted WSNs: Strong-Agent Coordinated Coverage and Vicsek-Driven...

Area of Science:

• Computer Science
• Robotics
• Wireless Communication

Background:

• Unmanned aerial vehicle (UAV) swarms are effective for data collection but face challenges with environmental dynamics and resource constraints.
• Existing centralized or pure reinforcement learning methods struggle to balance global solution quality with local flexibility in complex environments.

Purpose of the Study:

• To propose a hierarchical data collection framework for heterogeneous UAV-assisted wireless sensor networks (WSNs).
• To enhance coordination and decision-making for high-capability UAVs (H-UAVs) managing numerous low-capability UAVs (L-UAVs).
• To reduce the age of information (AoI) for ground users (GUs).

Main Methods:

• A hierarchical framework using multi-agent deep reinforcement learning (MADRL), specifically Multi-Agent Deep Deterministic Policy Gradient (MADDPG), for H-UAV coordination.
• Power-Voronoi partitioning dynamically updated based on GU data rates and L-UAV density.
• L-UAVs utilize a weighted Vicsek model for self-organized data collection and opportunistic relaying, considering AoI, link quality, and congestion.

Main Results:

• The proposed strong and weak agent MADDPG (SW-MADDPG) scheme demonstrated significant improvements in data collection efficiency.
• Reduced AoI by 30% compared to the No-Voronoi baseline.
• Reduced AoI by 21% compared to the Heuristic-HUAV baseline.

Conclusions:

• The hierarchical framework with spatial decomposition and decentralized weak-swarm control enables scalable and efficient data collection in large-scale UAV deployments.
• The SW-MADDPG approach effectively balances workload and adapts to environmental dynamics, outperforming existing methods in reducing AoI.