서브메뉴
검색
Resilient GPS Positioning Using Deep Neural Networks and Sensor Fusion With Factor Graph Optimization
Resilient GPS Positioning Using Deep Neural Networks and Sensor Fusion With Factor Graph Optimization
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20250211151405
- ISBN
- 9798382233307
- DDC
- 004
- 서명/저자
- Resilient GPS Positioning Using Deep Neural Networks and Sensor Fusion With Factor Graph Optimization
- 발행사항
- [Sl] : Stanford University, 2024
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2024
- 형태사항
- 134 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 85-11, Section: B.
- 주기사항
- Advisor: Grace Gao.
- 학위논문주기
- Thesis (Ph.D.)--Stanford University, 2024.
- 초록/해제
- 요약Autonomous vehicles (AVs), such as self driving cars and unmanned aerial vehicles, will operate in dense urban areas and require decimeter-level positioning estimates, both of which are challenging for traditional GNSS-based positioning algorithms, such as weighted least squares. More recent algorithms, such as methods using factor graph optimization (FGO) or deep neural networks (DNNs), can potentially satisfy this accuracy requirement of AVs in deep urban areas by leveraging the computational resources and multiple sensors available on these platforms. However, these algorithms are susceptible to GNSS signal vulnerabilities, like GNSS spoofing and faults both in received measurements and satellite states, which must be mitigated to ensure the accuracy and availability of position estimates from these algorithms. This dissertation describes contributions to this end and discusses methods that mitigate GPS spoofing attacks and measurement faults. First, we describe a DNN for GNSS-based positioning that is robust to measurement faults, such as additive biases. In our architecture, we solve challenges that emerge when applying traditional DNNs to the task of GNSS-based positioning using specialized architectures and by estimating corrections to initial positions. We validate our architecture on simulated and real-world measurements, showing that it has better accuracy than equivalent model-based approaches in the local Down direction, and can have better accuracy in the North and East direction depending on the initialization error in our method. In simulation, we also show that our approach effectively mitigates additive biases in measurements. Second, we discuss our method that uses switchable constraints (SC) in an FGO-based GNSS positioning algorithm to mitigate GPS spoofing attacks. We use odometry sensors to improve positioning accuracy and obtain measurements independent of GPS to mitigate spoofing. We validate our proposed method in simulation, showing that it effectively mitigates spoofing attacks while maintaining accuracy similar to that of a naive FGO under nominal conditions. Our method also incorporates the upcoming Chimera signal enhancement for "loop closure" to improve accuracy in the nominal case. Third, we discuss our modular expectation-maximization (EM)-based architecture that jointly mitigates GPS spoofing and measurement faults. We formulate the positioning problem as a two stage process which is iteratively solved using EM. In the first stage, we estimate the likelihoods that GNSS measurements are authentic and fault-free. In the second stage, we obtain position estimates, incorporating the likelihoods of authenticity and fault-free measurements to provide resilience to both spoofing attacks and measurements faults, whichever might be present. We validate our architecture with realistic simulated measurements, showing that it effectively mitigates faults and spoofing while maintaining accuracy similar to naive algorithms in nominal operating conditions. Fourth, we discuss gnss_lib_py, an open-source, modular, and extendable Python library for processing GNSS measurements, file types and datasets. gnss_lib_py also provides baseline implementations of traditional state estimation algorithms and methods to simulate realistic GNSS measurements. The methods that we describe in this dissertation enable spoofing and fault resilient GNSS positioning, enabling a safer future for the navigation of autonomous vehicles.
- 일반주제명
- Computer science
- 일반주제명
- Robotics
- 기타저자
- Stanford University.
- 기본자료저록
- Dissertations Abstracts International. 85-11B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008250123s2024 us c eng d■001000017161501
■00520250211151405
■006m o d
■007cr#unu||||||||
■020 ▼a9798382233307
■035 ▼a(MiAaPQ)AAI31255809
■035 ▼a(MiAaPQ)pd226jw5716
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a004
■1001 ▼aAshwin Vivek Kanhere.
■24510▼aResilient GPS Positioning Using Deep Neural Networks and Sensor Fusion With Factor Graph Optimization
■260 ▼a[Sl]▼bStanford University▼c2024
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2024
■300 ▼a134 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 85-11, Section: B.
■500 ▼aAdvisor: Grace Gao.
■5021 ▼aThesis (Ph.D.)--Stanford University, 2024.
■520 ▼aAutonomous vehicles (AVs), such as self driving cars and unmanned aerial vehicles, will operate in dense urban areas and require decimeter-level positioning estimates, both of which are challenging for traditional GNSS-based positioning algorithms, such as weighted least squares. More recent algorithms, such as methods using factor graph optimization (FGO) or deep neural networks (DNNs), can potentially satisfy this accuracy requirement of AVs in deep urban areas by leveraging the computational resources and multiple sensors available on these platforms. However, these algorithms are susceptible to GNSS signal vulnerabilities, like GNSS spoofing and faults both in received measurements and satellite states, which must be mitigated to ensure the accuracy and availability of position estimates from these algorithms. This dissertation describes contributions to this end and discusses methods that mitigate GPS spoofing attacks and measurement faults. First, we describe a DNN for GNSS-based positioning that is robust to measurement faults, such as additive biases. In our architecture, we solve challenges that emerge when applying traditional DNNs to the task of GNSS-based positioning using specialized architectures and by estimating corrections to initial positions. We validate our architecture on simulated and real-world measurements, showing that it has better accuracy than equivalent model-based approaches in the local Down direction, and can have better accuracy in the North and East direction depending on the initialization error in our method. In simulation, we also show that our approach effectively mitigates additive biases in measurements. Second, we discuss our method that uses switchable constraints (SC) in an FGO-based GNSS positioning algorithm to mitigate GPS spoofing attacks. We use odometry sensors to improve positioning accuracy and obtain measurements independent of GPS to mitigate spoofing. We validate our proposed method in simulation, showing that it effectively mitigates spoofing attacks while maintaining accuracy similar to that of a naive FGO under nominal conditions. Our method also incorporates the upcoming Chimera signal enhancement for "loop closure" to improve accuracy in the nominal case. Third, we discuss our modular expectation-maximization (EM)-based architecture that jointly mitigates GPS spoofing and measurement faults. We formulate the positioning problem as a two stage process which is iteratively solved using EM. In the first stage, we estimate the likelihoods that GNSS measurements are authentic and fault-free. In the second stage, we obtain position estimates, incorporating the likelihoods of authenticity and fault-free measurements to provide resilience to both spoofing attacks and measurements faults, whichever might be present. We validate our architecture with realistic simulated measurements, showing that it effectively mitigates faults and spoofing while maintaining accuracy similar to naive algorithms in nominal operating conditions. Fourth, we discuss gnss_lib_py, an open-source, modular, and extendable Python library for processing GNSS measurements, file types and datasets. gnss_lib_py also provides baseline implementations of traditional state estimation algorithms and methods to simulate realistic GNSS measurements. The methods that we describe in this dissertation enable spoofing and fault resilient GNSS positioning, enabling a safer future for the navigation of autonomous vehicles.
■590 ▼aSchool code: 0212.
■650 4▼aComputer science
■650 4▼aRobotics
■653 ▼aDeep neural networks
■653 ▼aFactor graph optimization
■653 ▼aAutonomous vehicles
■653 ▼aSwitchable constraints
■690 ▼a0984
■690 ▼a0771
■71020▼aStanford University.
■7730 ▼tDissertations Abstracts International▼g85-11B.
■790 ▼a0212
■791 ▼aPh.D.
■792 ▼a2024
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17161501▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


