BROWSE
Detail
Journal Article
Online and Safe Multi-Agent RL for Massive Random Access in Tactical Flying Ad-Hoc Networks
Cited - time in 
Share 
Share
- Authors
- Jimin Jeon, Jaeha Ahn, Min Lee, Youngbin You, Heejung Yu, Howon Lee
- Issue Date
- 2026-06
- Citation
- IEEE Internet of Things Journal, v.13, no.12, pp.25966-25979
- ISSN
- 2327-4662
- Publisher
- IEEE
- Language
- English
- Type
- Journal Article
- Abstract
- In massive random access-based tactical flying adhoc networks (FANETs), the rapid mobility of unmanned aerial vehicles (UAVs) leads to highly dynamic topological changes that frequently cause collisions of control and data packets and ultimately degrade network performance. To address the collision problem, this study proposes an online safe multi-agent reinforcement learning (OSMAR)-based massive random access method, incorporating an artificial Q-adjustment (AQA) mechanism to optimally allocate transmission slots to multiple UAVs within each frame. Specifically, a safe reinforcement learning (RL) technique is employed to design the action selection policy, which balances exploration and exploitation by considering long-term rewards and risks based on a Boltzmann distribution. With this, a blacklist mechanism is incorporated to prevent UAVs from choosing the time slots with high collision risk. Also, the AQA mechanism, composed of an artificial Q-decrement (AQD) that lowers the Q-values of slots already used by other UAVs and an artificial Q-initialization (AQI) that resets the Q-values of idle slots, leverages overheard channel status information to accelerate convergence and enhance adaptation under dynamic network conditions. Extensive simulations demonstrate that the proposed OSMAR method outperforms existing approaches in terms of collision probability, while also enhancing fairness and maintaining robustness under various network conditions.
- Keyword
- Massive random access, over-hearing mechanism, packet collisions, safe reinforcement learning, tactical FANET
- KSP Keywords
- Access method, Boltzmann distribution, Channel status, Collision Probability, Collision problem, Collision risk, Data packet, Dynamic Network, Existing Approaches, Exploration and exploitation, Flying Ad-hoc Networks