Learning-Enabled Adaptation and Planning for Space Robotics
Learning-Enabled Adaptation and Planning for Space Robotics
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20250211152744
- ISBN
- 9798342113083
- DDC
- 530
- 서명/저자
- Learning-Enabled Adaptation and Planning for Space Robotics
- 발행사항
- [Sl] : Stanford University, 2024
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2024
- 형태사항
- 134 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 86-04, Section: A.
- 주기사항
- Advisor: Pavone, Marco.
- 학위논문주기
- Thesis (Ph.D.)--Stanford University, 2024.
- 초록/해제
- 요약Space robots have always been a key component of space exploration, enabling humanity to explore distant and hazardous worlds. With advancements in machine learning and artificial intelligence, a new generation of "learning-enabled" robots is emerging, which are better suited to operating autonomously in unstructured, uncertain, and unforgiving environments. These robots could quickly and robustly make decisions about their own operation to maximize science objectives, navigate risky situations, and execute high-level instructions from human operators. The design of such learning-enabled algorithms for space robotics is the focus of this thesis.The first part of this thesis discusses using learning-based methods to accelerate both high-level and low-level planning in space robotics. We introduce a framework to improve the computational tractability of stochastic mission planning, unlocking the ability to plan optimally even in the presence of stochasticity, large state and action spaces, and long time horizons. We also introduce a method to speed up, i.e., "warm-start", trajectory optimization by using a learned model to generate good initialization trajectories, which lead to faster convergence.The second part of this thesis discusses using learning to adapt to novel and evolving environments. We show how this adaptation can be passive, by augmenting physics-based models with learned models, which are updated online as new environments are encountered, while preserving the interpretability of physical parameters. We also introduce a framework for active adaptation where the model monitors its own performance and curates a diverse subset of uncertain inputs to be used for periodic fine-tuning of the model, improving performance over the full data lifecycle. We conclude the thesis with a discussion of future research directions for space robotics.
- 일반주제명
- Physics
- 일반주제명
- Decision making
- 일반주제명
- Neural networks
- 일반주제명
- Labeling
- 일반주제명
- Robots
- 일반주제명
- Adaptation
- 일반주제명
- Distance learning
- 일반주제명
- Markov analysis
- 일반주제명
- Robotics
- 일반주제명
- Educational technology
- 기타저자
- Stanford University.
- 기본자료저록
- Dissertations Abstracts International. 86-04A.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008250123s2024 us c eng d■001000017163714
■00520250211152744
■006m o d
■007cr#unu||||||||
■020 ▼a9798342113083
■035 ▼a(MiAaPQ)AAI31520263
■035 ▼a(MiAaPQ)Stanforddm561sg6488
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a530
■1001 ▼aBanerjee, Somrita.
■24510▼aLearning-Enabled Adaptation and Planning for Space Robotics
■260 ▼a[Sl]▼bStanford University▼c2024
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2024
■300 ▼a134 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 86-04, Section: A.
■500 ▼aAdvisor: Pavone, Marco.
■5021 ▼aThesis (Ph.D.)--Stanford University, 2024.
■520 ▼aSpace robots have always been a key component of space exploration, enabling humanity to explore distant and hazardous worlds. With advancements in machine learning and artificial intelligence, a new generation of "learning-enabled" robots is emerging, which are better suited to operating autonomously in unstructured, uncertain, and unforgiving environments. These robots could quickly and robustly make decisions about their own operation to maximize science objectives, navigate risky situations, and execute high-level instructions from human operators. The design of such learning-enabled algorithms for space robotics is the focus of this thesis.The first part of this thesis discusses using learning-based methods to accelerate both high-level and low-level planning in space robotics. We introduce a framework to improve the computational tractability of stochastic mission planning, unlocking the ability to plan optimally even in the presence of stochasticity, large state and action spaces, and long time horizons. We also introduce a method to speed up, i.e., "warm-start", trajectory optimization by using a learned model to generate good initialization trajectories, which lead to faster convergence.The second part of this thesis discusses using learning to adapt to novel and evolving environments. We show how this adaptation can be passive, by augmenting physics-based models with learned models, which are updated online as new environments are encountered, while preserving the interpretability of physical parameters. We also introduce a framework for active adaptation where the model monitors its own performance and curates a diverse subset of uncertain inputs to be used for periodic fine-tuning of the model, improving performance over the full data lifecycle. We conclude the thesis with a discussion of future research directions for space robotics.
■590 ▼aSchool code: 0212.
■650 4▼aPhysics
■650 4▼aDecision making
■650 4▼aNeural networks
■650 4▼aLabeling
■650 4▼aRobots
■650 4▼aAdaptation
■650 4▼aDistance learning
■650 4▼aMarkov analysis
■650 4▼aRobotics
■650 4▼aEducational technology
■690 ▼a0771
■690 ▼a0605
■690 ▼a0800
■690 ▼a0710
■690 ▼a0796
■71020▼aStanford University.
■7730 ▼tDissertations Abstracts International▼g86-04A.
■790 ▼a0212
■791 ▼aPh.D.
■792 ▼a2024
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17163714▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.
