서브메뉴
검색
Real-Time Optimization for Robust State Estimation and Control of Legged Robots
Real-Time Optimization for Robust State Estimation and Control of Legged Robots
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20250211152817
- ISBN
- 9798384456278
- DDC
- 629.8
- 저자명
- Yang, Shuo.
- 서명/저자
- Real-Time Optimization for Robust State Estimation and Control of Legged Robots
- 발행사항
- [Sl] : Carnegie Mellon University, 2024
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2024
- 형태사항
- 238 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 86-03, Section: B.
- 주기사항
- Advisor: Manchester, Zachary.
- 학위논문주기
- Thesis (Ph.D.)--Carnegie Mellon University, 2024.
- 초록/해제
- 요약This thesis aims to provide methods and algorithms that enhance legged robot locomotion capabilities in various aspects. In recent years, more legged robot solutions have emerged and begun to assume real-world applications like construction site inspection and law enforcement. As legged robots enter unstructured real-world scenarios, they need improved motion control stability and more precise state estimation. Through methodical investigations and experiments, this research contributes several hardware and software innovations that demonstrate significantly improved stability, mobility, and autonomy for legged robots.On the control side, the thesis presents a legged robot hardware design incorporating two reaction wheels, which can be controlled alongside other joint motors in a model predictive controller. The additional reaction wheels greatly enhance the robot's stability. To harness the power of reaction wheels in real-time control, we explore the efficiency and flexibility of model predictive control. Due to the generic nature of its underlying numerical optimization framework, model predictive control can support different robot hardware designs within the same control framework.On the state estimation side, this thesis develops several real-time odometry solutions combining multiple inertial measurement units, joint encoders, contact sensors, and cameras to achieve low-drift position estimation during long-term locomotion. We first study two ways to model error sources in leg odometry to improve estimation performance. Then, we develop two visual-inertial leg odometry solutions that achieve state-of-the-art estimation accuracy. Along the way, we also systematically study Kalman filtering and factor graph-based optimization, which are crucial tools for general robot state estimation. Additionally, a chapter of this thesis is dedicated to the connection between optimization-based state estimation and optimization-based control through the factor graph. The factor graph, commonly used in large-scale state estimation and mapping, is a nuanced representation that explores sparsity in estimation problems. This similar sparsity, due to the inherent Markov property of robots, also appears in optimization-based trajectory generation and control. Thus, the graphical representation can illuminate some difficult control problems that are challenging to solve using conventional recursive methods.
- 일반주제명
- Robotics
- 일반주제명
- Engineering
- 일반주제명
- Computer science
- 일반주제명
- Mechanics
- 키워드
- Legged robots
- 키워드
- Optimization
- 키워드
- Sensor fusion
- 키워드
- SLAM
- 키워드
- State estimation
- 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■001000017163987
■00520250211152817
■006m o d
■007cr#unu||||||||
■020 ▼a9798384456278
■035 ▼a(MiAaPQ)AAI31558885
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a629.8
■1001 ▼aYang, Shuo.▼0(orcid)0000-0002-3291-9076
■24510▼aReal-Time Optimization for Robust State Estimation and Control of Legged Robots
■260 ▼a[Sl]▼bCarnegie Mellon University▼c2024
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2024
■300 ▼a238 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 86-03, Section: B.
■500 ▼aAdvisor: Manchester, Zachary.
■5021 ▼aThesis (Ph.D.)--Carnegie Mellon University, 2024.
■520 ▼aThis thesis aims to provide methods and algorithms that enhance legged robot locomotion capabilities in various aspects. In recent years, more legged robot solutions have emerged and begun to assume real-world applications like construction site inspection and law enforcement. As legged robots enter unstructured real-world scenarios, they need improved motion control stability and more precise state estimation. Through methodical investigations and experiments, this research contributes several hardware and software innovations that demonstrate significantly improved stability, mobility, and autonomy for legged robots.On the control side, the thesis presents a legged robot hardware design incorporating two reaction wheels, which can be controlled alongside other joint motors in a model predictive controller. The additional reaction wheels greatly enhance the robot's stability. To harness the power of reaction wheels in real-time control, we explore the efficiency and flexibility of model predictive control. Due to the generic nature of its underlying numerical optimization framework, model predictive control can support different robot hardware designs within the same control framework.On the state estimation side, this thesis develops several real-time odometry solutions combining multiple inertial measurement units, joint encoders, contact sensors, and cameras to achieve low-drift position estimation during long-term locomotion. We first study two ways to model error sources in leg odometry to improve estimation performance. Then, we develop two visual-inertial leg odometry solutions that achieve state-of-the-art estimation accuracy. Along the way, we also systematically study Kalman filtering and factor graph-based optimization, which are crucial tools for general robot state estimation. Additionally, a chapter of this thesis is dedicated to the connection between optimization-based state estimation and optimization-based control through the factor graph. The factor graph, commonly used in large-scale state estimation and mapping, is a nuanced representation that explores sparsity in estimation problems. This similar sparsity, due to the inherent Markov property of robots, also appears in optimization-based trajectory generation and control. Thus, the graphical representation can illuminate some difficult control problems that are challenging to solve using conventional recursive methods.
■590 ▼aSchool code: 0041.
■650 4▼aRobotics
■650 4▼aEngineering
■650 4▼aComputer science
■650 4▼aMechanics
■653 ▼aLegged robots
■653 ▼aModel predictive control
■653 ▼aOptimization
■653 ▼aSensor fusion
■653 ▼aSLAM
■653 ▼aState estimation
■690 ▼a0771
■690 ▼a0984
■690 ▼a0346
■690 ▼a0800
■690 ▼a0537
■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=T17163987▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.
