Optimizing underwater robotic technology using artificial intelligence

Alam, Shahbad (2025) Optimizing underwater robotic technology using artificial intelligence. Masters thesis, Memorial University of Newfoundland.

[English] PDF - Updated 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 (2MB)

Abstract

This thesis proposes to optimize the navigation of autonomous underwater vehicles by enhancing underwater acoustic communication and target sensing in complex oceanic environments. Current deep learning techniques in underwater acoustic communication often fail to account for domain knowledge, such as first-principles signal processing and acoustic propagation laws, which govern chaotic underwater environments. The main focus is on the integration of domain knowledge of the underwater environment in training neural networks. Such a context-aware deep learning model employs theory-trained neural network to accurately learn a non-linear map between input and output data. Details of the theory-trained neural network largely remains unexplored, with many open questions. This research introduces a context-aware methodology for regularizing neural networks. We explore three approaches to deep learning-based communication among AUVs. First, we advance long short-term memory (LSTM) neural networks for accurate underwater target detection that moves stealthily underwater. Second, we embed communication theory within a convolutional neural network (CNN) model in a supervised learning framework. Third, we test a minimally viable intelligent system that enables AUVs to communicate effectively in a hostile underwater environment. Through these venues, we provide a foundation for a more reliable and efficient underwater communication.

Actions (login required)

View Item