Navigation Support for an Autonomous Ferry Using Deep Reinforcement Learning in Simulated Maritime Environments

The development of shipping is witnessing increasing automation - from existing assistance systems to fully autonomous behaviour. In this article, we present a building block on the way to a fully autonomous passenger ferry for the Kiel Fjord in Germany by presenting a simulation-based approach to situation modelling of maritime environments and the behaviour therein. We show how this can be used for a posteriori analysis of the possible behaviour of the ship. This analysis then in turn flows into the decision-making process. We also use this environment to investigate the application of Deep Reinforcement Learning techniques to optimize navigation tasks and identify challenges and limitations.