본문

서브메뉴

Abstractions for Safety Assurance of Autonomous Systems
Abstractions for Safety Assurance of Autonomous Systems
Abstractions for Safety Assurance of Autonomous Systems

Detailed Information

자료유형  
 학위논문 서양
최종처리일시  
20260209102839
ISBN  
9798314843581
DDC  
004
저자명  
Hsieh, Chiao.
서명/저자  
Abstractions for Safety Assurance of Autonomous Systems
발행사항  
[Sl] : University of Illinois at Urbana-Champaign, 2023
발행사항  
Ann Arbor : ProQuest Dissertations & Theses, 2023
형태사항  
163 p
주기사항  
Source: Dissertations Abstracts International, Volume: 86-11, Section: B.
주기사항  
Advisor: Mitra, Sayan.
학위논문주기  
Thesis (Ph.D.)--University of Illinois at Urbana-Champaign, 2023.
초록/해제  
요약An autonomous system has components to perceive the environment, control the hardware, and communicate with other agents. Ideally, the formal analysis of the closed-loop system would have access to explicit models of all components. However, explicit models of perception and dynamics are often unavailable or intractable for formal verification. Instead, executable models for these components are available for simulation and testing in practice. In this thesis, we will present our compositional verification framework for certifying and assuring safety for systems with combinations of distributed communication, learning-enabled perception, and black-box dynamics. Our insight is to construct abstractions of these components for the end-to-end system-level safety analysis; then we empirically validate each component against its abstraction by sampling executable models.For distributed robotic applications, we focus on constructing the abstraction for heterogeneous motion dynamics. Our notion of port assumptions for dynamics decomposes the safety assurance into two steps: (a) the formal safety proof of the distributed system using port assumptions, and (b) the validation of port assumptions using data-driven reachability analyses. We learned that this compositional reasoning generalizes to both synchronous and asynchronous communications between heterogeneous vehicles. We are able to derive the collision avoidance guarantee in a distributed delivery application using our Koord framework for shared variable communication between ground vehicles and quadrotors. We apply the same idea on asynchronous message passing-based Unmanned Air-traffic Management protocols (UTM) and verify safe separations between quadrotors and fixed-wing airplanes.We also study autonomous systems with visual perception enabled by deep neural networks (DNNs). Our main insight is to search for ground truth-based approximate abstractions for perception. Our notion of approximate abstractions bypasses the challenges in the formal specification of perception and high dimensional image domains, and the precision of the approximate abstraction can be estimated empirically. We study both single agent and multiagent systems including three practical vision-based autonomous systems: (a) a lane tracking system for an autonomous vehicle, (b) a corn row following system for an agricultural robot, and (c) a vision-based multi-agent swarm formation. We are able to provide end-to-end assurance for all systems and examine the precision using simulations.In summary, our compositional verification framework outlines a pragmatic path to provide safety assurance of autonomous systems. We formalize (approximate) abstractions for compositional verification and develop approaches to search for abstractions. This allows us to formally verify as many components in the system as possible. For other components that are intractable for formal verification, we search for abstractions of these components and rigorously validate the abstractions via testing and data-driven verification. We are able to gain safety assurance using abstractions in all our case studies and validate the abstractions with high-fidelity Gazebo and AirSim simulations.
일반주제명  
Computer science
일반주제명  
Engineering
일반주제명  
Information technology
키워드  
Autonomous systems
키워드  
Safety assurance
키워드  
Formal verification
키워드  
Compositional verification
키워드  
Distributed robotics
키워드  
Learning-enabled systems
기타저자  
University of Illinois at Urbana-Champaign Computer Science
기본자료저록  
Dissertations Abstracts International. 86-11B.
전자적 위치 및 접속  
로그인 후 원문을 볼 수 있습니다.

MARC

 008260203s2023        us                              c    eng  d
■001000017365853
■00520260209102839
■006m          o    d                
■007cr#unu||||||||
■020    ▼a9798314843581
■035    ▼a(MiAaPQ)AAI32063785
■035    ▼a(MiAaPQ)httphdlhandlenet2142120127
■040    ▼aMiAaPQ▼cMiAaPQ
■0820  ▼a004
■1001  ▼aHsieh,  Chiao.
■24510▼aAbstractions  for  Safety  Assurance  of  Autonomous  Systems
■260    ▼a[Sl]▼bUniversity  of  Illinois  at  Urbana-Champaign▼c2023
■260  1▼aAnn  Arbor▼bProQuest  Dissertations  &  Theses▼c2023
■300    ▼a163  p
■500    ▼aSource:  Dissertations  Abstracts  International,  Volume:  86-11,  Section:  B.
■500    ▼aAdvisor:  Mitra,  Sayan.
■5021  ▼aThesis  (Ph.D.)--University  of  Illinois  at  Urbana-Champaign,  2023.
■520    ▼aAn  autonomous  system  has  components  to  perceive  the  environment,  control  the  hardware,  and  communicate  with  other  agents.  Ideally,  the  formal  analysis  of  the  closed-loop  system  would  have  access  to  explicit  models  of  all  components.  However,  explicit  models  of  perception  and  dynamics  are  often  unavailable  or  intractable  for  formal  verification.  Instead,  executable  models  for  these  components  are  available  for  simulation  and  testing  in  practice.  In  this  thesis,  we  will  present  our  compositional  verification  framework  for  certifying  and  assuring  safety  for  systems  with  combinations  of  distributed  communication,  learning-enabled  perception,  and  black-box  dynamics.  Our  insight  is  to  construct  abstractions  of  these  components  for  the  end-to-end  system-level  safety  analysis;  then  we  empirically  validate  each  component  against  its  abstraction  by  sampling  executable  models.For  distributed  robotic  applications,  we  focus  on  constructing  the  abstraction  for  heterogeneous  motion  dynamics.  Our  notion  of  port  assumptions  for  dynamics  decomposes  the  safety  assurance  into  two  steps:  (a)  the  formal  safety  proof  of  the  distributed  system  using  port  assumptions,  and  (b)  the  validation  of  port  assumptions  using  data-driven  reachability  analyses.  We  learned  that  this  compositional  reasoning  generalizes  to  both  synchronous  and  asynchronous  communications  between  heterogeneous  vehicles.  We  are  able  to  derive  the  collision  avoidance  guarantee  in  a  distributed  delivery  application  using  our  Koord  framework  for  shared  variable  communication  between  ground  vehicles  and  quadrotors.  We  apply  the  same  idea  on  asynchronous  message  passing-based  Unmanned  Air-traffic  Management  protocols  (UTM)  and  verify  safe  separations  between  quadrotors  and  fixed-wing  airplanes.We  also  study  autonomous  systems  with  visual  perception  enabled  by  deep  neural  networks  (DNNs).  Our  main  insight  is  to  search  for  ground  truth-based  approximate  abstractions  for  perception.  Our  notion  of  approximate  abstractions  bypasses  the  challenges  in  the  formal  specification  of  perception  and  high  dimensional  image  domains,  and  the  precision  of  the  approximate  abstraction  can  be  estimated  empirically.  We  study  both  single  agent  and  multiagent  systems  including  three  practical  vision-based  autonomous  systems:  (a)  a  lane  tracking  system  for  an  autonomous  vehicle,  (b)  a  corn  row  following  system  for  an  agricultural  robot,  and  (c)  a  vision-based  multi-agent  swarm  formation.  We  are  able  to  provide  end-to-end  assurance  for  all  systems  and  examine  the  precision  using  simulations.In  summary,  our  compositional  verification  framework  outlines  a  pragmatic  path  to  provide  safety  assurance  of  autonomous  systems.  We  formalize  (approximate)  abstractions  for  compositional  verification  and  develop  approaches  to  search  for  abstractions.  This  allows  us  to  formally  verify  as  many  components  in  the  system  as  possible.  For  other  components  that  are  intractable  for  formal  verification,  we  search  for  abstractions  of  these  components  and  rigorously  validate  the  abstractions  via  testing  and  data-driven  verification.  We  are  able  to  gain  safety  assurance  using  abstractions  in  all  our  case  studies  and  validate  the  abstractions  with  high-fidelity  Gazebo  and  AirSim  simulations.
■590    ▼aSchool  code:  0090.
■650  4▼aComputer  science
■650  4▼aEngineering
■650  4▼aInformation  technology
■653    ▼aAutonomous  systems
■653    ▼aSafety  assurance
■653    ▼aFormal  verification
■653    ▼aCompositional  verification
■653    ▼aDistributed  robotics
■653    ▼aLearning-enabled  systems
■690    ▼a0984
■690    ▼a0489
■690    ▼a0800
■690    ▼a0537
■71020▼aUniversity  of  Illinois  at  Urbana-Champaign▼bComputer  Science.
■7730  ▼tDissertations  Abstracts  International▼g86-11B.
■790    ▼a0090
■791    ▼aPh.D.
■792    ▼a2023
■793    ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17365853▼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

    Подробнее информация.

    • Бронирование
    • не существует
    • моя папка
    • Первый запрос зрения
    • Non-Book Loan Application
    • Nighttime Book Loan Application
    материал
    Reg No. Количество платежных Местоположение статус Ленд информации
    TF17068 전자도서 대출가능 My Folder 부재도서신고 비도서대출신청 야간 도서대출신청

    * Бронирование доступны в заимствований книги. Чтобы сделать предварительный заказ, пожалуйста, нажмите кнопку бронирование

    Books borrowed together with this book

    Related Popular Books

    Available after logging in.