Oceanscape: A graph-based framework for autonomous coastal navigation

This paper presents Oceanscape, a novel graph-based framework designed to advance the development and research of decision support systems for vessels, as well as algorithms for pathfinding and autonomous navigation. The framework addresses key limitations of existing maritime data representations by providing a unified geometric structure that bridges traditional navigation systems with modern machine learning approaches. Oceanscape uses geospatial triangulations to construct a detailed graph representation of the marine environment, integrating diverse data products including coastline polygons, depth data, sea marks, and fairway systems. This approach enables both traditional maritime applications and geometric deep learning research through compatibility with modern frameworks, while supporting multi-agent scenarios and situational awareness through separate network layers. The framework's efficacy is demonstrated through a detailed case study set in the Oslofjord in Norway, where it is employed to develop a system that generates port-to-port routes within the region, accounting for the presence of sea marks and fairway systems. The case study highlights the effectiveness of the proposed graph structures in handling intricate coastline geometries, integrating environmental factors, and optimizing navigational paths.