Answers are grounded in this paper and cite their sources.
Efficient and adaptive routing remains a critical challenge in dynamic network environments due to changing traffic conditions, network congestion, and link failures. Traditional routing approaches, such as Dijkstra and OSPF, rely on static or semi-adaptive mechanisms, which limit their ability to respond to real-time network variations. To address these limitations, this paper proposes an Agentic Multi-Agent Reinforcement Learning (AMARL) framework for intelligent and decentralized routing. In the proposed model, each network node operates as an autonomous agent capable of observing local network conditions, communicating with neighboring agents, and making cooperative routing decisions. A multi-objective reward function is designed to jointly optimize delay, throughput, packet delivery ratio, and packet loss. The framework enables dynamic adaptation to network changes without requiring global recomputation. Extensive simulations are conducted under varying network sizes, traffic conditions, and link failure scenarios. The results demonstrate that the proposed AMARL model achieves significant performance improvements, including up to 45% higher throughput, 35% lower delay, and over 50% reduction in packet loss compared to traditional methods. Furthermore, the model maintains higher reliability under link failures and exhibits fast convergence with reduced energy consumption over training episodes. These findings indicate that the integration of agentic intelligence and multi-agent cooperation provides a scalable, adaptive, and efficient solution for next-generation network routing. The proposed framework is particularly suitable for dynamic and large-scale network environments where traditional routing methods fail to perform effectively.
No takes yet. Share an insight, caveat, or question.
Nur et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: