A study on aiming control and validation of a reinforcement learning-based fire monitor in maritime environments

Abstract

With the growing interest in autonomous ships, there has been an increasing demand for intelligent fire suppression systems that can operate reliably even under irregular sea conditions. This study proposes an autonomous aiming control approach for a fire monitor, designed to optimize its targeting performance through reinforcement learning (RL). A virtual environment was created using the Unity physics engine, which simulated ship motion and the dynamic behavior of discharged water while reflecting variations in sea states and discharge pressure. The proposed RL-based model was trained to improve aiming efficiency by observing the vessel's motion and nozzle pressure data measured by an inertial measurement unit (IMU) sensor and pressure sensor. The effectiveness of the model was verified through full-scale experiments conducted on an actual vessel operating under sea state about 3 condition. Experimental observations confirmed that the system maintained stable and precise fire-targeting performance in a real maritime environment. These results suggest that the proposed RL-based control approach can contribute to the automation of fire response operations for future unmanned autonomous ships.