서브메뉴
검색
Contact Interactions for Robot Dexterity
Contact Interactions for Robot Dexterity
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20250211152725
- ISBN
- 9798384089667
- DDC
- 629.8
- 저자명
- Cheng, Xianyi.
- 서명/저자
- Contact Interactions for Robot Dexterity
- 발행사항
- [Sl] : Carnegie Mellon University, 2024
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2024
- 형태사항
- 194 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 86-03, Section: B.
- 주기사항
- Advisor: Mason, Matthew T.
- 학위논문주기
- Thesis (Ph.D.)--Carnegie Mellon University, 2024.
- 초록/해제
- 요약We have remarkable AIs and precise robots, but why do humans still do so much tedious physical work, from factories to households? A core missing piece is robot dexterity. Human-level dexterity is vastly complex, manifesting in versatile motions to clean and sort delicate dishes, reactive control to assemble tiny electronic parts, and high adaptability for any unseen scenarios. True dexterity is a complex synergy of mind and body and the crucial intelligence of task-solving and execution. Achieving human-level dexterity in robotic manipulation will transform industries and people's lives, empowering robots to perform complex manual processes as seamlessly as humans in factories, households, service industries, space operations, etc. This thesis discusses the key aspect of robot dexterity --- contact interactions. Contacts introduce discontinuities into the robot systems, creating a complex hybrid and high-dimensional space. There are many ways for the robot, the object, and the environment to make contact with each other, making seemingly innumerable dexterous manipulation skills. This thesis explores a new perspective --- using auto-enumerated local contact modes as auto-generated motion primitives. Our efficient contact mode enumeration algorithm finds all kinematically feasible contact modes in milliseconds. Contact modes can represent diverse types of local contact interactions. Each contact mode corresponds to a set of continuous constrained dynamical equations, which can guide fast and smooth motion integration. I combined the contact mode guidance with planning algorithms, which globally explore long-horizon contact-rich solutions: from 2D to 3D scenarios with sampling-based planning algorithms and then with an optimized Monte-Carlo tree search structure and hierarchical contact strategy optimization. My algorithms are the first to plan for a variety of novel dexterous manipulation tasks in seconds, tested on multiple robot platforms and over 20 task types, requiring little tuning for new tasks. This thesis demonstrates that contact mode-based skills offer three benefits: versatility, generalizability, and robustness. For versatility, contact modes are complete in capturing all local contact interactions. Contact modes help generate more diverse manipulation motions and require less engineering than using manually defined primitives. For generalizability, contact mode enumeration handles any rigid contact configuration with first-order approximation, inherently accommodating many scenarios for rigid bodies, even those with irregular, non-convex shapes. For robustness, when combined with force and compliance control, we have skills that are robust against uncertainty. This thesis presents experiments on different robots, most of which are low DoF devices, including a vacuum suction cup, a point end-effector, a parallel gripper, a dexterous direct-drive hand, and an array of low-cost delta robots. I intend these experiments to motivate one idea --- dexterity is a capability all robots can and should develop in the future.
- 일반주제명
- Robotics
- 일반주제명
- Mechanical engineering
- 일반주제명
- Computer engineering
- 기타저자
- Carnegie Mellon University Mechanical Engineering
- 기본자료저록
- Dissertations Abstracts International. 86-03B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008250123s2024 us c eng d■001000017163568
■00520250211152725
■006m o d
■007cr#unu||||||||
■020 ▼a9798384089667
■035 ▼a(MiAaPQ)AAI31490126
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a629.8
■1001 ▼aCheng, Xianyi.▼0(orcid)0000-0001-8342-9459
■24510▼aContact Interactions for Robot Dexterity
■260 ▼a[Sl]▼bCarnegie Mellon University▼c2024
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2024
■300 ▼a194 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 86-03, Section: B.
■500 ▼aAdvisor: Mason, Matthew T.
■5021 ▼aThesis (Ph.D.)--Carnegie Mellon University, 2024.
■520 ▼aWe have remarkable AIs and precise robots, but why do humans still do so much tedious physical work, from factories to households? A core missing piece is robot dexterity. Human-level dexterity is vastly complex, manifesting in versatile motions to clean and sort delicate dishes, reactive control to assemble tiny electronic parts, and high adaptability for any unseen scenarios. True dexterity is a complex synergy of mind and body and the crucial intelligence of task-solving and execution. Achieving human-level dexterity in robotic manipulation will transform industries and people's lives, empowering robots to perform complex manual processes as seamlessly as humans in factories, households, service industries, space operations, etc. This thesis discusses the key aspect of robot dexterity --- contact interactions. Contacts introduce discontinuities into the robot systems, creating a complex hybrid and high-dimensional space. There are many ways for the robot, the object, and the environment to make contact with each other, making seemingly innumerable dexterous manipulation skills. This thesis explores a new perspective --- using auto-enumerated local contact modes as auto-generated motion primitives. Our efficient contact mode enumeration algorithm finds all kinematically feasible contact modes in milliseconds. Contact modes can represent diverse types of local contact interactions. Each contact mode corresponds to a set of continuous constrained dynamical equations, which can guide fast and smooth motion integration. I combined the contact mode guidance with planning algorithms, which globally explore long-horizon contact-rich solutions: from 2D to 3D scenarios with sampling-based planning algorithms and then with an optimized Monte-Carlo tree search structure and hierarchical contact strategy optimization. My algorithms are the first to plan for a variety of novel dexterous manipulation tasks in seconds, tested on multiple robot platforms and over 20 task types, requiring little tuning for new tasks. This thesis demonstrates that contact mode-based skills offer three benefits: versatility, generalizability, and robustness. For versatility, contact modes are complete in capturing all local contact interactions. Contact modes help generate more diverse manipulation motions and require less engineering than using manually defined primitives. For generalizability, contact mode enumeration handles any rigid contact configuration with first-order approximation, inherently accommodating many scenarios for rigid bodies, even those with irregular, non-convex shapes. For robustness, when combined with force and compliance control, we have skills that are robust against uncertainty. This thesis presents experiments on different robots, most of which are low DoF devices, including a vacuum suction cup, a point end-effector, a parallel gripper, a dexterous direct-drive hand, and an array of low-cost delta robots. I intend these experiments to motivate one idea --- dexterity is a capability all robots can and should develop in the future.
■590 ▼aSchool code: 0041.
■650 4▼aRobotics
■650 4▼aMechanical engineering
■650 4▼aComputer engineering
■653 ▼aContact mechanics
■653 ▼aDexterous manipulation
■653 ▼aRobotic manipulation
■653 ▼aLocal contact modes
■653 ▼aManipulation tasks
■690 ▼a0771
■690 ▼a0548
■690 ▼a0464
■71020▼aCarnegie Mellon University▼bMechanical Engineering.
■7730 ▼tDissertations Abstracts International▼g86-03B.
■790 ▼a0041
■791 ▼aPh.D.
■792 ▼a2024
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17163568▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.
