Reinforcement Learning Optimizes Sensor Selection for Maritime Vessel Tracking

A study published on arXiv (2607.22667) presents a reinforcement learning strategy designed for sensor selection in diverse maritime networks to monitor a single ship. This method employs information-gain-guided sensor management, where a developed policy picks one relevant sensor for tracking during each decision interval, thus eliminating the need for activating all sensors or conducting online evaluations of expected information gain. To estimate the vessel's state from noisy data, a Bayesian sequential Monte Carlo tracker is utilized, offering a belief representation suitable for nonlinear and non-Gaussian scenarios. A Proximal Policy Optimization agent operates within a georeferenced simulation of the CMMI Smart Marina testbed located at Ayia Napa Marina, Cyprus, assessing features such as belief state, detection history, coverage, sensor geometry, and realized information gain, with rewards based on tracking precision.