본문

서브메뉴

Generative Reconstruction of Human Motion
Generative Reconstruction of Human Motion
Generative Reconstruction of Human Motion

Detailed Information

자료유형  
 학위논문 서양
최종처리일시  
20260202105613
ISBN  
9798265429438
DDC  
006.3
저자명  
Li, Jiaman.
서명/저자  
Generative Reconstruction of Human Motion
발행사항  
[Sl] : Stanford University, 2025
발행사항  
Ann Arbor : ProQuest Dissertations & Theses, 2025
형태사항  
111 p
주기사항  
Source: Dissertations Abstracts International, Volume: 87-06, Section: B.
주기사항  
Advisor: Liu, Karen;Wu, Jiajun.
학위논문주기  
Thesis (Ph.D.)--Stanford University, 2025.
초록/해제  
요약Human movement is inherently complex and diverse, grounded in interaction with the physical world-ranging from everyday actions like walking or cooking to skilled motions such as dance or gymnastics. These motions are essential for enabling embodied agents, whether virtual avatars or physical robots, to move purposefully and interact effectively. In virtual environments, human motion drives realistic character behavior, supports robot policy training in simulation, and powers lifelike avatars in mixed-reality applications. In the physical world, humanoid robots can benefit from human motion data to learn agile motor skills-like dancing-and accomplish functional tasks such as preparing coffee or performing household chores. While motion capture provides the most accurate way to obtain human motion data, it requires specialized hardware, controlled environments, and time-consuming setup, which limits scalability and diversity. As a result, capturing the breadth of human activities found in everyday life remains difficult. To overcome these limitations, we can turn to more accessible sources-data that are easier to capture or already widely available, such as third-person videos, egocentric recordings, or object-centric signals from a single sensor. These modalities offer the potential to recover diverse human motions across a broad range of real-world scenarios, but they are inherently sparse and under-constrained.In this dissertation, I address the challenge of reconstructing human motion from such sparse and under-constrained observations by framing the problem within a generative modeling paradigm. I develop a series of generative approaches that leverage learned motion priors to reconstruct 3D human motion from 2D video or other minimal inputs. First, I introduce a framework that leverages 2D motion priors learned from monocular video to reconstruct generalized 3D motion. Second, I show how egocentric video can be used to infer full-body motion grounded in 3D environments. Third, I present a method for reconstructing human motion from dynamic object trajectories. Finally, I extend this formulation to handle even sparser input by generating human and object motion from abstract control signals such as language. Together, these contributions enable scalable and data-efficient acquisition of human motion from accessible observations, broaden the diversity of motions that can be generated, and move us closer to creating intelligent agents that act and interact naturally in the physical world.
일반주제명  
Diffusion models
일반주제명  
Geometry
일반주제명  
Robotics
키워드  
Gymnastics
키워드  
Physical robots
기타저자  
Stanford University.
기본자료저록  
Dissertations Abstracts International. 87-06B.
전자적 위치 및 접속  
로그인 후 원문을 볼 수 있습니다.

MARC

 008260126s2025        us                              c    eng  d
■001000017360746
■00520260202105613
■006m          o    d                
■007cr#unu||||||||
■020    ▼a9798265429438
■035    ▼a(MiAaPQ)AAI32316429
■035    ▼a(MiAaPQ)Stanfordsp112tf7344
■040    ▼aMiAaPQ▼cMiAaPQ
■0820  ▼a006.3
■1001  ▼aLi,  Jiaman.
■24510▼aGenerative  Reconstruction  of  Human  Motion
■260    ▼a[Sl]▼bStanford  University▼c2025
■260  1▼aAnn  Arbor▼bProQuest  Dissertations  &  Theses▼c2025
■300    ▼a111  p
■500    ▼aSource:  Dissertations  Abstracts  International,  Volume:  87-06,  Section:  B.
■500    ▼aAdvisor:  Liu,  Karen;Wu,  Jiajun.
■5021  ▼aThesis  (Ph.D.)--Stanford  University,  2025.
■520    ▼aHuman  movement  is  inherently  complex  and  diverse,  grounded  in  interaction  with  the  physical  world-ranging  from  everyday  actions  like  walking  or  cooking  to  skilled  motions  such  as  dance  or  gymnastics.  These  motions  are  essential  for  enabling  embodied  agents,  whether  virtual  avatars  or  physical  robots,  to  move  purposefully  and  interact  effectively.  In  virtual  environments,  human  motion  drives  realistic  character  behavior,  supports  robot  policy  training  in  simulation,  and  powers  lifelike  avatars  in  mixed-reality  applications.  In  the  physical  world,  humanoid  robots  can  benefit  from  human  motion  data  to  learn  agile  motor  skills-like  dancing-and  accomplish  functional  tasks  such  as  preparing  coffee  or  performing  household  chores.  While  motion  capture  provides  the  most  accurate  way  to  obtain  human  motion  data,  it  requires  specialized  hardware,  controlled  environments,  and  time-consuming  setup,  which  limits  scalability  and  diversity.  As  a  result,  capturing  the  breadth  of  human  activities  found  in  everyday  life  remains  difficult.  To  overcome  these  limitations,  we  can  turn  to  more  accessible  sources-data  that  are  easier  to  capture  or  already  widely  available,  such  as  third-person  videos,  egocentric  recordings,  or  object-centric  signals  from  a  single  sensor.  These  modalities  offer  the  potential  to  recover  diverse  human  motions  across  a  broad  range  of  real-world  scenarios,  but  they  are  inherently  sparse  and  under-constrained.In  this  dissertation,  I  address  the  challenge  of  reconstructing  human  motion  from  such  sparse  and  under-constrained  observations  by  framing  the  problem  within  a  generative  modeling  paradigm.  I  develop  a  series  of  generative  approaches  that  leverage  learned  motion  priors  to  reconstruct  3D  human  motion  from  2D  video  or  other  minimal  inputs.  First,  I  introduce  a  framework  that  leverages  2D  motion  priors  learned  from  monocular  video  to  reconstruct  generalized  3D  motion.  Second,  I  show  how  egocentric  video  can  be  used  to  infer  full-body  motion  grounded  in  3D  environments.  Third,  I  present  a  method  for  reconstructing  human  motion  from  dynamic  object  trajectories.  Finally,  I  extend  this  formulation  to  handle  even  sparser  input  by  generating  human  and  object  motion  from  abstract  control  signals  such  as  language.  Together,  these  contributions  enable  scalable  and  data-efficient  acquisition  of  human  motion  from  accessible  observations,  broaden  the  diversity  of  motions  that  can  be  generated,  and  move  us  closer  to  creating  intelligent  agents  that  act  and  interact  naturally  in  the  physical  world.
■590    ▼aSchool  code:  0212.
■650  4▼aDiffusion  models
■650  4▼aGeometry
■650  4▼aRobotics
■653    ▼aGymnastics
■653    ▼aPhysical  robots
■690    ▼a0771
■71020▼aStanford  University.
■7730  ▼tDissertations  Abstracts  International▼g87-06B.
■790    ▼a0212
■791    ▼aPh.D.
■792    ▼a2025
■793    ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17360746▼nKERIS▼z이  자료의  원문은  한국교육학술정보원에서  제공합니다.

Preview

Export

ChatGPT Discussion

AI Recommended Related Books


    New Books MORE
    Statistics for the past 3 years. Go to brief

    detalle info

    • Reserva
    • No existe
    • Mi carpeta
    • Primera solicitud
    • Non-Book Loan Application
    • Nighttime Book Loan Application
    Material
    número de libro número de llamada Ubicación estado Prestar info
    TF17679 전자도서 대출가능 My Folder 부재도서신고 비도서대출신청 야간 도서대출신청

    * Las reservas están disponibles en el libro de préstamos. Para hacer reservaciones, haga clic en el botón de reserva

    Books borrowed together with this book

    Related Popular Books

    Available after logging in.