Perceive, Predict, and Plan: Robotic Expeditionary Science in Oceanic Spatiotemporal Fields
An improved understanding of our ocean would allow us to characterize the largest habitable biosphere on planet Earth, quantify the geochemical processes that control Earth’s climate, and develop responsible regulations for controlling the natural resources stored in its depths. Expeditionary scienc...
Main Author: | |
---|---|
Other Authors: | |
Format: | Thesis |
Published: |
Massachusetts Institute of Technology
2023
|
Online Access: | https://hdl.handle.net/1721.1/150717 https://orcid.org/0000-0002-5249-2866 |
_version_ | 1811083718759022592 |
---|---|
author | Preston, Victoria Lynn |
author2 | Roy, Nicholas |
author_facet | Roy, Nicholas Preston, Victoria Lynn |
author_sort | Preston, Victoria Lynn |
collection | MIT |
description | An improved understanding of our ocean would allow us to characterize the largest habitable biosphere on planet Earth, quantify the geochemical processes that control Earth’s climate, and develop responsible regulations for controlling the natural resources stored in its depths. Expeditionary science is the art of collecting in situ observations of an environment to build approximate models of underlying properties that move us towards this understanding. Robotic platforms are a critical technology for collecting observations of the ocean. Depth-capable autonomous underwater vehicles (AUVs) are commonly used to build static maps of the seafloor by executing pre-programmed surveys. However, there is growing urgency to generate rich data products of spatiotemporal distributions that characterize the physics and chemistry of the deep ocean biogeosphere. In this thesis, the problem of charting dynamic deep sea hydrothermal plumes with depth-capable AUVs is investigated. Effectively collecting samples of geochemical plumes using the operationally preferred strategy of pre-specifying surveys requires access to a dynamics model of the advective currents, bathymetric updrafts, and turbulent mixing at a hydrothermal site. In practice, however, access to this information is unavailable, imperfect, or only partially known, and so a model of plume dynamics must be inferred from observations and subsequently leveraged to improve future sampling performance. As most in situ scientific instruments yield point-measurements, considerable uncertainty is placed over the form of the dynamics in purely data-driven solutions. Challenges related to planning under uncertainty for geochemical surveys in the deep ocean are addressed in this thesis by embedding scientific knowledge as a strong inductive prior for tractable model learning and decision-making. Algorithmic contributions of this thesis show how plumes can be perceived from field data, their fate predicted far into the future (e.g., multiple days), and informative fixed trajectories planned which place an AUV in the right place at the right time. Scientific assessment of observational data collected with AUV Sentry during field trials in the Guaymas Basin, Gulf of California are interwoven with algorithmic analyses, demonstrating how intelligent perception, prediction, and planning enables novel insights about hydrothermal plumes. |
first_indexed | 2024-09-23T12:38:01Z |
format | Thesis |
id | mit-1721.1/150717 |
institution | Massachusetts Institute of Technology |
last_indexed | 2024-09-23T12:38:01Z |
publishDate | 2023 |
publisher | Massachusetts Institute of Technology |
record_format | dspace |
spelling | mit-1721.1/1507172023-05-16T03:28:57Z Perceive, Predict, and Plan: Robotic Expeditionary Science in Oceanic Spatiotemporal Fields Preston, Victoria Lynn Roy, Nicholas Michel, Anna P. M. Massachusetts Institute of Technology. Department of Aeronautics and Astronautics Joint Program in Oceanography/Applied Ocean Science and Engineering An improved understanding of our ocean would allow us to characterize the largest habitable biosphere on planet Earth, quantify the geochemical processes that control Earth’s climate, and develop responsible regulations for controlling the natural resources stored in its depths. Expeditionary science is the art of collecting in situ observations of an environment to build approximate models of underlying properties that move us towards this understanding. Robotic platforms are a critical technology for collecting observations of the ocean. Depth-capable autonomous underwater vehicles (AUVs) are commonly used to build static maps of the seafloor by executing pre-programmed surveys. However, there is growing urgency to generate rich data products of spatiotemporal distributions that characterize the physics and chemistry of the deep ocean biogeosphere. In this thesis, the problem of charting dynamic deep sea hydrothermal plumes with depth-capable AUVs is investigated. Effectively collecting samples of geochemical plumes using the operationally preferred strategy of pre-specifying surveys requires access to a dynamics model of the advective currents, bathymetric updrafts, and turbulent mixing at a hydrothermal site. In practice, however, access to this information is unavailable, imperfect, or only partially known, and so a model of plume dynamics must be inferred from observations and subsequently leveraged to improve future sampling performance. As most in situ scientific instruments yield point-measurements, considerable uncertainty is placed over the form of the dynamics in purely data-driven solutions. Challenges related to planning under uncertainty for geochemical surveys in the deep ocean are addressed in this thesis by embedding scientific knowledge as a strong inductive prior for tractable model learning and decision-making. Algorithmic contributions of this thesis show how plumes can be perceived from field data, their fate predicted far into the future (e.g., multiple days), and informative fixed trajectories planned which place an AUV in the right place at the right time. Scientific assessment of observational data collected with AUV Sentry during field trials in the Guaymas Basin, Gulf of California are interwoven with algorithmic analyses, demonstrating how intelligent perception, prediction, and planning enables novel insights about hydrothermal plumes. Ph.D. Ph.D. 2023-05-15T19:34:41Z 2023-05-15T19:34:41Z 2023-02 2023-02-15T14:05:42.822Z Thesis https://hdl.handle.net/1721.1/150717 https://orcid.org/0000-0002-5249-2866 In Copyright - Educational Use Permitted Copyright retained by author(s) https://rightsstatements.org/page/InC-EDU/1.0/ application/pdf Massachusetts Institute of Technology |
spellingShingle | Preston, Victoria Lynn Perceive, Predict, and Plan: Robotic Expeditionary Science in Oceanic Spatiotemporal Fields |
title | Perceive, Predict, and Plan: Robotic Expeditionary Science in Oceanic Spatiotemporal Fields |
title_full | Perceive, Predict, and Plan: Robotic Expeditionary Science in Oceanic Spatiotemporal Fields |
title_fullStr | Perceive, Predict, and Plan: Robotic Expeditionary Science in Oceanic Spatiotemporal Fields |
title_full_unstemmed | Perceive, Predict, and Plan: Robotic Expeditionary Science in Oceanic Spatiotemporal Fields |
title_short | Perceive, Predict, and Plan: Robotic Expeditionary Science in Oceanic Spatiotemporal Fields |
title_sort | perceive predict and plan robotic expeditionary science in oceanic spatiotemporal fields |
url | https://hdl.handle.net/1721.1/150717 https://orcid.org/0000-0002-5249-2866 |
work_keys_str_mv | AT prestonvictorialynn perceivepredictandplanroboticexpeditionaryscienceinoceanicspatiotemporalfields |