Learning Maritime Surface Ship by Imitation Learning
Landais, Clément (2021)
Landais, Clément
2021
All rights reserved. This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi-fe2021061738661
https://urn.fi/URN:NBN:fi-fe2021061738661
Tiivistelmä
Autonomy is being developed further and further in the maritime navigation area. The wish to create such autonomous vessels has several logistic, economic and environmental origins. Safety can be improved by applying automatic adequate responses to risk-prone situations. Efficiency in the ship’s route, which implies both fewer costs and a smaller carbon footprint, can be optimized with computer-level precision. The issue is to develop sufficiently reliable solutions for this critical area.
This master’s thesis aims at giving an answer to this problem by using Imitation Learning. The goal is to build an autonomous agent that performs well in surface maritime navigation. The algorithmic approach is based on three main steps. The first one consists of isolating the best human behaviours from a provided dataset. The second one is to identify, in this reduced "expert" dataset, which navigation principles and rules are followed. Finally, the agent learns to imitate these expert actions. This agent is implemented in Python and assessed in a simulated maritime environment.
This master’s thesis aims at giving an answer to this problem by using Imitation Learning. The goal is to build an autonomous agent that performs well in surface maritime navigation. The algorithmic approach is based on three main steps. The first one consists of isolating the best human behaviours from a provided dataset. The second one is to identify, in this reduced "expert" dataset, which navigation principles and rules are followed. Finally, the agent learns to imitate these expert actions. This agent is implemented in Python and assessed in a simulated maritime environment.