서브메뉴
검색
Adversarial Scenario Generation for Virtual Testing of Off-Road Autonomous Vehicles
Adversarial Scenario Generation for Virtual Testing of Off-Road Autonomous Vehicles
Detailed Information
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202105245
- ISBN
- 9798291569726
- DDC
- 621
- 저자명
- Sender, Ted.
- 서명/저자
- Adversarial Scenario Generation for Virtual Testing of Off-Road Autonomous Vehicles
- 발행사항
- [Sl] : University of Michigan, 2025
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2025
- 형태사항
- 115 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-02, Section: B.
- 주기사항
- Advisor: Epureanu, Bogdan I.
- 학위논문주기
- Thesis (Ph.D.)--University of Michigan, 2025.
- 초록/해제
- 요약Developing autonomous vehicles (AVs) that operate in diverse and demanding environments is a difficult challenge. These vehicles naturally have advanced vehicle dynamics, and to rise to the environment's demand, they consequently are controlled by sophisticated autonomy systems, which often consist of interconnected perception, navigation, and control systems. Understanding the weaknesses and limitations of these complex systems is crucial for understanding what they can and cannot do safely, properly, or at all. Simulation has become a valuable tool for AV testing because it is cheaper, easier, and safer than physical testing. This work focuses on autonomous ground vehicles, and the two classes of such vehicles are on-road/urban AVs and off-road AVs. On-road AVs operate in highly structured environments, but must deal with the uncertainty of what the surrounding dynamic actors (vehicles, pedestrians, etc.) are doing. On the other hand, off-road AVs must operate across a wide variety of highly unstructured and mostly static environments that consist of rugged (and potentially hazardous) terrain whose ground surface is often made of multiple non-uniform materials (e.g., soft-deformable soil, water, foliage, etc.). Much of the existing work for AV development and testing pertains to on-road/urban AVs and is thus centered around the challenges that plague on-road AVs (i.e., uncertainty of the dynamic actors). However, the off-road AV community is slowly growing and is deserving of similarly advanced tools to meet their development and testing needs. This dissertation focuses on developing one of the most important types of testing tools for off-road AVs, an area that is mostly non-existent. We specifically focus on the creation of an adversarial algorithm that can efficiently create scenarios that highlight weaknesses or limitations in the autonomy system. We propose a method called Black-Box Adversarially Compounding Regret Through Evolution (BACRE), which is an evolutionary algorithm guided by a novel regret-based metric for general navigation tasks. A black-box approach is often preferable when system complexity can be diverse, like with off-road AVs, and when whole-system testing is required. A custom simulation platform is also provided to assist with the automated testing of AVs in unstructured environments. We start by presenting a base version of BACRE and perform numerical experiments to demonstrate its potential. We then address several of BACRE's limitations and extend it to the case of generating adversarial 3-dimensional terrains, resulting in the BACRE-3D algorithm. The numerical experiments show that BACRE-3D is capable of pushing the entire simulation platform to its limits; BACRE-3D finds failure cases that show the chosen simulation platform is incapable of properly modeling scenarios it was expected to be capable of modeling, and it finds scenarios that push a state-of-the-art off-road trajectory planner to its limits. In summary, this dissertation fills a void in the (meager) off-road AV testing literature by showing how to break down the intractable problem of finding performance limitations of off-road AVs in high-dimensional, static, unstructured environments.
- 일반주제명
- Mechanical engineering
- 일반주제명
- Computer engineering
- 일반주제명
- Automotive engineering
- 키워드
- Physical testing
- 기타저자
- University of Michigan Mechanical Engineering
- 기본자료저록
- Dissertations Abstracts International. 87-02B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2025 us c eng d■001000017359979
■00520260202105245
■006m o d
■007cr#unu||||||||
■020 ▼a9798291569726
■035 ▼a(MiAaPQ)AAI32272040
■035 ▼a(MiAaPQ)umichrackham006501
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a621
■1001 ▼aSender, Ted.
■24510▼aAdversarial Scenario Generation for Virtual Testing of Off-Road Autonomous Vehicles
■260 ▼a[Sl]▼bUniversity of Michigan▼c2025
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2025
■300 ▼a115 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-02, Section: B.
■500 ▼aAdvisor: Epureanu, Bogdan I.
■5021 ▼aThesis (Ph.D.)--University of Michigan, 2025.
■520 ▼aDeveloping autonomous vehicles (AVs) that operate in diverse and demanding environments is a difficult challenge. These vehicles naturally have advanced vehicle dynamics, and to rise to the environment's demand, they consequently are controlled by sophisticated autonomy systems, which often consist of interconnected perception, navigation, and control systems. Understanding the weaknesses and limitations of these complex systems is crucial for understanding what they can and cannot do safely, properly, or at all. Simulation has become a valuable tool for AV testing because it is cheaper, easier, and safer than physical testing. This work focuses on autonomous ground vehicles, and the two classes of such vehicles are on-road/urban AVs and off-road AVs. On-road AVs operate in highly structured environments, but must deal with the uncertainty of what the surrounding dynamic actors (vehicles, pedestrians, etc.) are doing. On the other hand, off-road AVs must operate across a wide variety of highly unstructured and mostly static environments that consist of rugged (and potentially hazardous) terrain whose ground surface is often made of multiple non-uniform materials (e.g., soft-deformable soil, water, foliage, etc.). Much of the existing work for AV development and testing pertains to on-road/urban AVs and is thus centered around the challenges that plague on-road AVs (i.e., uncertainty of the dynamic actors). However, the off-road AV community is slowly growing and is deserving of similarly advanced tools to meet their development and testing needs. This dissertation focuses on developing one of the most important types of testing tools for off-road AVs, an area that is mostly non-existent. We specifically focus on the creation of an adversarial algorithm that can efficiently create scenarios that highlight weaknesses or limitations in the autonomy system. We propose a method called Black-Box Adversarially Compounding Regret Through Evolution (BACRE), which is an evolutionary algorithm guided by a novel regret-based metric for general navigation tasks. A black-box approach is often preferable when system complexity can be diverse, like with off-road AVs, and when whole-system testing is required. A custom simulation platform is also provided to assist with the automated testing of AVs in unstructured environments. We start by presenting a base version of BACRE and perform numerical experiments to demonstrate its potential. We then address several of BACRE's limitations and extend it to the case of generating adversarial 3-dimensional terrains, resulting in the BACRE-3D algorithm. The numerical experiments show that BACRE-3D is capable of pushing the entire simulation platform to its limits; BACRE-3D finds failure cases that show the chosen simulation platform is incapable of properly modeling scenarios it was expected to be capable of modeling, and it finds scenarios that push a state-of-the-art off-road trajectory planner to its limits. In summary, this dissertation fills a void in the (meager) off-road AV testing literature by showing how to break down the intractable problem of finding performance limitations of off-road AVs in high-dimensional, static, unstructured environments.
■590 ▼aSchool code: 0127.
■650 4▼aMechanical engineering
■650 4▼aComputer engineering
■650 4▼aAutomotive engineering
■653 ▼aAutonomous vehicles
■653 ▼aAdversarial scenario generation
■653 ▼aPhysical testing
■653 ▼aEvolutionary algorithm
■690 ▼a0548
■690 ▼a0464
■690 ▼a0540
■71020▼aUniversity of Michigan▼bMechanical Engineering.
■7730 ▼tDissertations Abstracts International▼g87-02B.
■790 ▼a0127
■791 ▼aPh.D.
■792 ▼a2025
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17359979▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.
Preview
Export
ChatGPT Discussion
AI Recommended Related Books
Подробнее информация.
- Бронирование
- не существует
- моя папка
- Первый запрос зрения
- Non-Book Loan Application
- Nighttime Book Loan Application
Available after logging in.


