Towards Intelligent and Adaptive Multi-Agent Slicing in Maritime Networks

Integrated aerial-terrestrial maritime networks represent communication-constrained environments that often suffer from limited adaptability and interoperability. To overcome these limitations, we design an intelligent network slicing framework leveraging Open Radio Access Network (O-RAN) features. We propose an asymmetric-state hierarchical multiagent reinforcement learning algorithm to achieve adaptive quality of service-aware maritime connectivity. By hierarchically structuring discrete and continuous agents, the algorithm handles the Virtual Network Function (VNF) orchestration actions, the RAN slicing actions, and their interdependency. The two agents are coordinated via a novel inter-agent communication module based on various proposed approaches designed to study the communication between the agents. The results demonstrate that the proposed solution outperforms single-agent benchmarks, achieving about 65% higher reward than the discrete baseline and over 3x the reward of the continuous benchmark. It also highlights the key role of the inter-agent communication module in improving the agents' coordination and enhancing the overall performance.