Leveraging Big Data and ML to Aid Planning of Maritime Activities

In the maritime domain, the planning of operations typically aims at optimizing over basic cost and efficiency metrics. While these metrics are used efficiently for simpler tasks-like inventory management and route planning-in more complex scenarios -like large-scale surveying operations-more operational parameters need to be integrated in the decision process. This paper presents a proof of concept of a Decision-Making Aid (DMA) which adopts a human-machine teaming approach using a machine learning algorithm designed to assist a human operator by generating a set of optimal plans derived from historical data. Specifically, the paper focuses on using the tool to assist in the scheduling of underwater survey vehicles, aiming to maximize both the quality of the sonar data collected and the coverage of the area of interest. Leveraging the increasing amount of open-source data available, the algorithm takes into consideration the spatial and time-varying nature of the operational parameters. We present several experimental results involving a mixture of simulation techniques, discretization, and reinforcement learning to demonstrate the benefits of using machine learning to assist maritime planning operations. Our results show how the DMA can enable Tasking Authorities to work in more complex environments and make informed decisions in favour of maximizing value-survey quality and coverage. Our tool can be used to efficiently simulate and plan long-term operations involving multiple assets.