Multi-Modal Inter-Domain Routing Based on Multi-Agent Reinforcement Learning on Heterogeneous Networks

To address the increasing complexity of heterogeneous networks, especially for applications such as telemedicine and autonomous driving, this study introduces a novel operational framework for efficient multi-modal, multi-objective inter-domain routing. While existing research has largely focused on multi-objective inter-domain routing, significant challenges remain in meeting multi-modal demands. Moreover, reinforcement learning-based routing algorithms often encounter issues with slow convergence and limited efficiency. This framework combines a multi-modal dung beetle optimizer with a multi-agent, multi-modal deep reinforcement learning (DRL) algorithm. The dung beetle optimizer mitigates the convergence and efficiency limitations of current routing algorithms by creating a high-dimensional action space, which the DRL algorithm leverages to enhance routing performance. Experimental results show a 30% increase in throughput, 25% reduction in latency, and 20% decrease in packet loss, highlighting the framework's potential for advancing multimodal routing.