Reinforcement learning based ship autopilot for transit in ice

Dawson, Michael (2022) Reinforcement learning based ship autopilot for transit in ice. Masters thesis, Memorial University of Newfoundland.

[img] [English] PDF - Accepted Version
Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission.

Download (6MB)


Aided navigation systems are constantly evolving and lending themselves to a growing number of applications. With a range of benefits from improved safety to reduced costs, they are an important concept within the shipping industry. This work aims to improve safety at sea in northern regions through demonstrating the potential for a ship autopilot specifically designed for transit in ice. Autopilots on ships are not a new concept – PID (proportional-integral-derivative) controller-based autopilots are already implemented onboard vessels. PID controllers, however, do not perform well when dealing with ice floes. This is because they are designed to maintain a process – such as ship speed and heading – without proper information regarding the ice around the vessel. Instead, this work aims to implement an adaptive autopilot that adjusts based on ice information ahead of the ship. The autopilot is a reinforcement learning based model that is trained using proximal policy optimization. Within a simulated setting, the ship autopilot is shown to be able to safely navigate through ice under controlled conditions. Both the ability to travel through ice at reasonable speeds as well as avoid individual ice floes is demonstrated. Due to the limited generalizability of the current method, further development is necessary before being applied in real world scenarios.

Item Type: Thesis (Masters)
Item ID: 15638
Additional Information: Includes bibliographical references (pages 58-64)
Keywords: machine learning, reinforcement learning, ship autopilot, safety at sea, marine simulation, ice loads
Department(s): Engineering and Applied Science, Faculty of
Date: October 2022
Date Type: Submission
Digital Object Identifier (DOI):
Library of Congress Subject Heading: Machine learning; Reinforcement learning; Automatic pilot (Ships); Ships--Safety measures; Marine engineering--Simulation methods; Loads (Mechanics)

Actions (login required)

View Item View Item


Downloads per month over the past year

View more statistics