Optimizing Ocean Feature Estimation and Tracking through Adaptive Sampling and Formation Control of Autonomous Underwater Vehicles
Time: Thu 2023-09-28 10.00
Location: Kollegiesalen, Brinellvägen 8, Stockholm
Subject area: Electrical Engineering
Doctoral student: Joana Filipa Gouveia Fonseca , Reglerteknik
Opponent: Associate Professor Nina Mahmoudian,
Supervisor: Professor Karl H. Johansson, Reglerteknik
Increased ocean temperatures caused by climate change are expected to lead to more frequent and severe harmful algal blooms, which deteriorate water quality, cause human illness and fish mortality. Scientific understanding of algal blooms and their dynamics is limited due to the lack of data from such ocean phenomena. State-of-the-art ocean monitoring includes satellite imagery and dedicated research vessels. Mobile sensors based on autonomous underwater vehicles (AUVs) and other robotic technologies are of growing importance for efficient environmental monitoring of the oceans. The overall objective of this thesis is to design a system for ocean feature estimation and tracking based on adaptive sensor sampling using AUVs. The thesis contributions are focused on the following three problems.
The first problem we consider is how to estimate and track circular and non-circular ocean features using a multi-robot system. We propose a circumnavigation control law, proving that it forces the AUVs to converge to a circular formation. Two target estimation algorithms are presented: one is based on a leader-follower approach, while the other is distributed. Both algorithms are shown to successfully estimate and track the mobile target's location. Secondly, we consider the problem of tracking ocean fronts using a single AUV supported by satellite data. We develop a Gaussian process model for the front estimation and show how it can be updated based on the available sensor and satellite data. Using this model, a control law is developed that guides the AUV to move toward and along the ocean front. The closed-loop system is evaluated through a detailed simulation environment with realistic vehicle and environment models and real algal bloom data. Finally, we develop an experimental setup based on a real AUV to demonstrate that our method for algal bloom tracking is feasible in practice. We show experimental results from two surveys in the Stockholm archipelago and compare the performance of the real system with simulation studies. The results indicate that the front tracking and gradient estimation algorithms are working well but also suggest important items for further studies.