서브메뉴
검색
Towards Safe and Aligned Embodied AI in the Era of Robotics Foundation Models
Towards Safe and Aligned Embodied AI in the Era of Robotics Foundation Models
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202105111
- ISBN
- 9798293893362
- DDC
- 629.8
- 저자명
- Tian, Thomas.
- 서명/저자
- Towards Safe and Aligned Embodied AI in the Era of Robotics Foundation Models
- 발행사항
- [Sl] : University of California, Berkeley, 2025
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2025
- 형태사항
- 119 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-04, Section: B.
- 주기사항
- Advisor: Tomizuka, Masayoshi;Bajcsy, Andrea.
- 학위논문주기
- Thesis (Ph.D.)--University of California, Berkeley, 2025.
- 초록/해제
- 요약The success of Large Language Models (LLMs) has transformed many areas of non- embodied artificial intelligence, from language understanding and generation to multimodal reasoning. Their ability to learn from internet-scale data and generalize across diverse tasks has raised the following question in the robotics community: Can we replicatethis success with robotics foundation models-large-scale models that directly generate robot actions from sensory input?Indeed, robotics foundation models pre-trained on internet-scale data are beginning to reshape how robots understand the complex world, interpret human feedback, and plan actions. These models hold the promise of significantly improving generalization, enabling robots to operate reliably in increasingly unstructured, dynamic, and novel real-world scenarios.However, despite the remarkable progress, it is precisely this integration of foundation models that introduces new safety and alignment challenges in robotics. Robots are safety-critical systems, wherein a foundation model's single erroneous visual or language interpretation, misaligned behavior generation, or high inference latency can lead to catastrophic consequences. As robotics models become more capable, ensuring that their actions are safe and aligned with human values becomes increasingly urgent.In an era where both academic and industrial researchers are racing to scale up robotics models, this dissertation takes a step back, offering a timely investigation into the safety and alignment challenges that arise across the life cycle of robotics foundation models, from pre-training strategies, to post-training alignment, to ensuring safety at deployment.
- 일반주제명
- Robotics
- 일반주제명
- Computer science
- 키워드
- Robots
- 키워드
- Sensory input
- 기타저자
- University of California, Berkeley Mechanical Engineering
- 기본자료저록
- Dissertations Abstracts International. 87-04B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2025 us c eng d■001000017359378
■00520260202105111
■006m o d
■007cr#unu||||||||
■020 ▼a9798293893362
■035 ▼a(MiAaPQ)AAI32237037
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a629.8
■1001 ▼aTian, Thomas.
■24510▼aTowards Safe and Aligned Embodied AI in the Era of Robotics Foundation Models
■260 ▼a[Sl]▼bUniversity of California, Berkeley▼c2025
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2025
■300 ▼a119 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-04, Section: B.
■500 ▼aAdvisor: Tomizuka, Masayoshi;Bajcsy, Andrea.
■5021 ▼aThesis (Ph.D.)--University of California, Berkeley, 2025.
■520 ▼aThe success of Large Language Models (LLMs) has transformed many areas of non- embodied artificial intelligence, from language understanding and generation to multimodal reasoning. Their ability to learn from internet-scale data and generalize across diverse tasks has raised the following question in the robotics community: Can we replicatethis success with robotics foundation models-large-scale models that directly generate robot actions from sensory input?Indeed, robotics foundation models pre-trained on internet-scale data are beginning to reshape how robots understand the complex world, interpret human feedback, and plan actions. These models hold the promise of significantly improving generalization, enabling robots to operate reliably in increasingly unstructured, dynamic, and novel real-world scenarios.However, despite the remarkable progress, it is precisely this integration of foundation models that introduces new safety and alignment challenges in robotics. Robots are safety-critical systems, wherein a foundation model's single erroneous visual or language interpretation, misaligned behavior generation, or high inference latency can lead to catastrophic consequences. As robotics models become more capable, ensuring that their actions are safe and aligned with human values becomes increasingly urgent.In an era where both academic and industrial researchers are racing to scale up robotics models, this dissertation takes a step back, offering a timely investigation into the safety and alignment challenges that arise across the life cycle of robotics foundation models, from pre-training strategies, to post-training alignment, to ensuring safety at deployment.
■590 ▼aSchool code: 0028.
■650 4▼aRobotics
■650 4▼aComputer science
■653 ▼aLarge Language Models
■653 ▼aRobots
■653 ▼aFoundation models
■653 ▼aSensory input
■653 ▼aLanguage interpretation
■690 ▼a0771
■690 ▼a0800
■690 ▼a0984
■71020▼aUniversity of California, Berkeley▼bMechanical Engineering.
■7730 ▼tDissertations Abstracts International▼g87-04B.
■790 ▼a0028
■791 ▼aPh.D.
■792 ▼a2025
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17359378▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


