DC-HEN: A Deadline-aware and Congestion-relieved Hierarchical Emergency Navigation Algorithm for Ship Indoor Environments

Abstract

Emergency evacuation is critical following a ship accident, as passengers are required to escape the dynamic hazards and reach the muster station before the deadline. In the existing efforts, users are guided to a safe path away from the danger, but unconstrained detours may mislead users to miss the ship capsizing deadline. Another major drawback is the heavy congestion during crowd evacuation. Therefore, this paper proposes DC-HEN, a hierarchical emergency navigation algorithm with both deadline and congestion awareness for ship indoor environments. Taking advantage of reinforcement learning techniques, DC-HEN can provide an individually customized evacuation route for each user in a real-time manner. We validate the proposed approach in a large-scale simulation environment with different population sizes based on a real-ship indoor scenario. Compared with the state-of-the-art solutions (CANS, ECSSN), experimental results show that DC-HEN can trade off between path efficiency and congestion to guide users to the exit safely.