If autonomous navigation in dense harbours and waterways will be made possible, knowledge on different domains will have to be incorporated. The paper presents a short background on model-based and model-free track controllers that are used for systematic research on the accessibility of ships in shallow and confined water. Both control methods were evaluated based on a simulation of a 110 m inland vessel at different loading conditions when navigating on the river Seine in Paris at different water levels, discharges and wind conditions. The model-based controller gives far more precise control in comparison with human controlled tracks on the same simulator platform than model-free controllers. Therefore, an investigation started on the use of multi-paradigm simulation methods that combine different training and validation platforms for model predictive reinforcement learning in, for example, path following. The principle of this methodology is shown in a simulation on the Western Scheldt and will be further investigated in the DDSHIP, Data-driven smart shipping, project.
Eloot et al. (Wed,) studied this question.