An Abruption Routing Decision Algorithm Towards UAV Network Based on Online Adaptive Decision Replacement Mechanism-enabled MARL

• Cuimei Liu,
• Zhenyu Wang,
• Xin Zhang,
• Chi Jia,
• Chuan Lin,
• Guangjie Han,
• Yating Zhang,

Vol. 20, No. 1, January 31, 2026
10.3837/tiis.2026.01.010, Download Paper (Free):

• UAV Network
• Multi-Agent Reinforcement Learning
• Reward estimation and recalibration mechanism
• Online adaptive decision replacement mechanism
• MADDPG

Abstract

This paper proposes advanced mechanisms for autonomous routing decision-making in Unmanned Aerial Vehicle (UAV) networks. Ensuring seamless data transmission in dynamic UAV networks faces key challenges like multi-hop routing delays and frequent routing abruptions. Existing approaches struggle to precisely model multi-hop transmission dynamics or achieve rapid recovery from routing failures. On account of Multi-Agent Reinforcement Learning (MARL) algorithm, this paper proposes the Reward Estimation and Recalibration (RER) Mechanism. By implementing a path efficiency evaluation and spatio-temporal decay allocation strategy, RER rectifies the bias in long-term value estimation. For routing abruption, the Online Adaptive Decision Replacement (OADR) mechanism is developed, by slicing MARL stages and introducing traditional routing protocols for temporary decision-making to suppress the cascading effects of routing disruptions. Based on RER and OADR mechanisms, the RROA-MADDPG algorithm is proposed for abrusption routing decision for UAV network. Experiments show, the proposed RROA-MADDPG achieves a 8% higher data delivery rate than traditional methods, delivering data to the sink node within 60 hops with delay below 0.05 ms under dynamic link conditions and routing abruptions.

Statistics

Cite this article