Navigation Situation Adaptive Learning-Based Path Planning of Maritime Autonomous Surface Ships

Abstract

In this paper, a navigation situation adaptive learning-based path planning (NSAL-PP) scheme is created for a maritime autonomous surface ships (MASS) with a hierarchical deep reinforcement learning (HDRL) algorithm. In the first level of hierarchy, the MASS navigational situation is adaptively learnt from the ontology theory and the Protégé logical language in terms of entities and attributes. In the second level of hierarchy, path planning skills are learnt by combining with deep Q-learning, the environment model, ship behavior space, reward function and exploration and utilization strategy. Specifically, the reward function consists of safety and navigational task. Finally, the simulations are built in the Python and 2D-Pygame platform, with Tianjin Port of China as a case study. Both simulation and experimental results demonstrate that the proposed NSAL-PP method is feasible and the collision free navigation is achieved, especially for narrow channel (waterway).