Rörelseprediktion för omkringliggande marina farkoster

There is continuous development in the field of autonomous vessels, particularly within maritime environments. Among the key challenges is the ability to predict the behavior and maneuvers of surrounding surface vessels in various navigational scenarios. This thesis, conducted in collaboration with ABB, focuses on the development of a trajectory prediction algorithm designed for use in congested and narrow maritime environments. The proposed algorithm aims to handle complex scenarios involving potential collision risks, specifically Give-Way, Head-On, and Overtaking situations. Additionally, it seeks to estimate the long-term destination of nearby vessels to improve the accuracy of trajectory prediction. To validate the performance of the algorithm, a series of custom-designed simulation scenarios were implemented. These scenarios adhered to the COLREGs (International Regulations for Preventing Collisions at Sea) and varied in complexity, combining multiple types of collision risks to create dense and hazardous environments. The study demonstrates that a context-aware prediction model will outperform a basic constant velocity model, particularly in the context of collision avoidance. Moreover, the results highlight the importance of accurate COLREGs situation classification, reliable goal point estimation, and robust maneuver modeling. These components collectively influence the quality of the predicted trajectories. Further results and analysis of the algorithm's performance are presented and discussed in the result.