Multi-Agent Distributed Cooperative Routing for Maritime Emergency Communication

With the development of intelligent shipping industry, massive terminal device is connected to maritime network, which causes the centralized scheduling mechanism fails to meet the communication requirements of large-scale network. Mean-while, constant location changes of vessels make it difficult to acquire the optimal route planning with existing routing schemes. Therefore, enlightened by the success of multi-agent reinforcement learning (MARL), we propose a multi-agent distributed cooperative routing algorithm driven by maritime emergency communication task. The algorithm utilizes the calculation results of adjacent agents to train local model, which can alleviate the coupling between individual agent and global data. For large network with small-scale topological changes, we leverage online learning mechanism for local training to ensure the accuracy of routing decision. The simulation results demonstrate that the proposed routing algorithm could not only avoid congestion, but substantially reduce retraining time, overhead communication cost and high compute consumption brought by small-scale topological dynamic changes. The corresponding open-source repository is shared on Github.