Development of trajectory-tracking maneuvering system for automatic berthing/unberthing based on double deep Q-network and experimental validation with an actual large ferry

Abstract

The development of an automatic berthing and unberthing system is essential for realizing autonomous ship navigation systems. Although numerous berthing/unberthing methods using neural networks (NNs) have been proposed in recent years, whether NNs trained via numerical simulations including modeling errors can be used for the automatic berthing/unberthing of actual ships in natural external disturbances is not clear. Hence, in this study, a maneuvering system for automatic berthing/unberthing was developed using a NN that solves the trajectory-tracking problem. The proposed method considered future targets as discrete points in learning, such as to absorb the modeling error and reduce the influence of external disturbances; thus, it achieved a more robust control compared to path planning methods. The system achieved trajectory tracking at low speeds by offline learning without teaching data to follow a target ship with the same maneuverability as its own ship and operated randomly under external disturbances. Furthermore, an online target correction logic function that locally corrects targets was developed to increase the tracking ability in external disturbances. Subsequently, a berthing and unberthing experiment was conducted using an actual large ferry to ensure the applicability of simulation-based NNs to the trajectory-tracking control of an actual ship in a real-life environment.