서브메뉴
검색
Understanding the Human-Automated Vehicle Interaction and Its Traffic-Level Implications
Understanding the Human-Automated Vehicle Interaction and Its Traffic-Level Implications
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202105132
- ISBN
- 9798291559901
- DDC
- 385
- 저자명
- Zhong, Xinzhi.
- 서명/저자
- Understanding the Human-Automated Vehicle Interaction and Its Traffic-Level Implications
- 발행사항
- [Sl] : The University of Wisconsin - Madison, 2025
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2025
- 형태사항
- 114 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-02, Section: B.
- 주기사항
- Advisor: Ahn, Soyoung.
- 학위논문주기
- Thesis (Ph.D.)--The University of Wisconsin - Madison, 2025.
- 초록/해제
- 요약Vehicles on the road today have various automation features considered SAE Level 2-4, largely proprietary to individual automakers. No uniform standards for these automation features currently exist. This can give rise to highly heterogeneous and mixed traffic, consisting of automated vehicles (AVs) with potentially wide-varying behaviors by design and human-driven vehicles (HDVs) with highly variable behaviors by nature. The control logic embedded in the automation systems also appears to diverge significantly from human driver preference, evidenced by frequent voluntary interventions initiated by users, not the automation systems. Such control transitions could spell trouble for traffic performance, instigating or intensifying traffic flow instability. Yet, there is a limited understanding of human-AV interactions in dynamic environments, posing a major challenge to advancing the design of shared-control AV systems.This thesis aims to understand the human-AV interactions in car-following and their system-level implications for traffic flow, and to leverage these insights to advance human-AV collaboration in shared-control driving systems. Specific objectives are to: (1) understand the heterogeneity of commercial AVs through a uniform analysis platform; (2) characterize voluntary human driver interventions and their impacts on both individual vehicles and traffic flow; (3) advance the design of automation control systems by incorporating multiple objectives, including mitigating the undesirable impacts of voluntary interventions; and (4) enhance the interpretability of human-AV interactive driving through human-aligned reasoning. To this end, we introduce a unifying stochastic framework to understand the heterogeneity of AVs compared to HDVs, with a particular emphasis on the connection between driving behaviors and their traffic-level impacts. The findings signify the variability (1) between AVs and HDVs, and (2) across AV developers, engine modes, and speed ranges, albeit to a lesser degree than HDV behaviors. It illuminates the limitations of the commercial AVs in system-level performances in terms of traffic hysteresis, a phenomenon closely related to traffic stability. Through driving simulator-based experiments, we investigate the adverse effects of voluntary interventions on traffic stability, particularly disturbance propagation. We find that the deviation from human preferences in automated driving is a critical factor prompting voluntary interventions. By decoding the decision-making process for interventions with an evidence accumulation model, we develop an AV car-following control strategy based on deep reinforcement learning to effectively minimize unnecessary interventions and improve traffic stability. Further, to promote intuitive human understanding of human-AV interactions, we propose LISA (large language models (LLMs)-integrated synthetic human agents), a multi-modal generative model designed to reason and interpret control transition through natural language. This framework reinforces the interpretability of human-AV interactions by integrating AV behavioral pattern inference with human cognitive reasoning, and by characterizing their influence on traffic throughput and hysteresis. Built upon LISA, an adaptive collaborative driving algorithm is developed to facilitate seamless handling of control transitions, ultimately improving traffic throughput and mitigating traffic hysteresis.
- 일반주제명
- Transportation
- 일반주제명
- Automotive engineering
- 키워드
- Generative model
- 기타저자
- The University of Wisconsin - Madison Civil & Environmental Engr
- 기본자료저록
- Dissertations Abstracts International. 87-02B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2025 us c eng d■001000017359525
■00520260202105132
■006m o d
■007cr#unu||||||||
■020 ▼a9798291559901
■035 ▼a(MiAaPQ)AAI32239276
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a385
■1001 ▼aZhong, Xinzhi.
■24510▼aUnderstanding the Human-Automated Vehicle Interaction and Its Traffic-Level Implications
■260 ▼a[Sl]▼bThe University of Wisconsin - Madison▼c2025
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2025
■300 ▼a114 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-02, Section: B.
■500 ▼aAdvisor: Ahn, Soyoung.
■5021 ▼aThesis (Ph.D.)--The University of Wisconsin - Madison, 2025.
■520 ▼aVehicles on the road today have various automation features considered SAE Level 2-4, largely proprietary to individual automakers. No uniform standards for these automation features currently exist. This can give rise to highly heterogeneous and mixed traffic, consisting of automated vehicles (AVs) with potentially wide-varying behaviors by design and human-driven vehicles (HDVs) with highly variable behaviors by nature. The control logic embedded in the automation systems also appears to diverge significantly from human driver preference, evidenced by frequent voluntary interventions initiated by users, not the automation systems. Such control transitions could spell trouble for traffic performance, instigating or intensifying traffic flow instability. Yet, there is a limited understanding of human-AV interactions in dynamic environments, posing a major challenge to advancing the design of shared-control AV systems.This thesis aims to understand the human-AV interactions in car-following and their system-level implications for traffic flow, and to leverage these insights to advance human-AV collaboration in shared-control driving systems. Specific objectives are to: (1) understand the heterogeneity of commercial AVs through a uniform analysis platform; (2) characterize voluntary human driver interventions and their impacts on both individual vehicles and traffic flow; (3) advance the design of automation control systems by incorporating multiple objectives, including mitigating the undesirable impacts of voluntary interventions; and (4) enhance the interpretability of human-AV interactive driving through human-aligned reasoning. To this end, we introduce a unifying stochastic framework to understand the heterogeneity of AVs compared to HDVs, with a particular emphasis on the connection between driving behaviors and their traffic-level impacts. The findings signify the variability (1) between AVs and HDVs, and (2) across AV developers, engine modes, and speed ranges, albeit to a lesser degree than HDV behaviors. It illuminates the limitations of the commercial AVs in system-level performances in terms of traffic hysteresis, a phenomenon closely related to traffic stability. Through driving simulator-based experiments, we investigate the adverse effects of voluntary interventions on traffic stability, particularly disturbance propagation. We find that the deviation from human preferences in automated driving is a critical factor prompting voluntary interventions. By decoding the decision-making process for interventions with an evidence accumulation model, we develop an AV car-following control strategy based on deep reinforcement learning to effectively minimize unnecessary interventions and improve traffic stability. Further, to promote intuitive human understanding of human-AV interactions, we propose LISA (large language models (LLMs)-integrated synthetic human agents), a multi-modal generative model designed to reason and interpret control transition through natural language. This framework reinforces the interpretability of human-AV interactions by integrating AV behavioral pattern inference with human cognitive reasoning, and by characterizing their influence on traffic throughput and hysteresis. Built upon LISA, an adaptive collaborative driving algorithm is developed to facilitate seamless handling of control transitions, ultimately improving traffic throughput and mitigating traffic hysteresis.
■590 ▼aSchool code: 0262.
■650 4▼aTransportation
■650 4▼aAutomotive engineering
■650 4▼aEnvironmental engineering
■653 ▼aAutomated vehicles
■653 ▼aDriver interventions
■653 ▼aGenerative model
■653 ▼aHuman-AV interactions
■653 ▼aLongitudinal control
■653 ▼aTraffic disturbance
■690 ▼a0709
■690 ▼a0543
■690 ▼a0775
■690 ▼a0540
■71020▼aThe University of Wisconsin - Madison▼bCivil & Environmental Engr.
■7730 ▼tDissertations Abstracts International▼g87-02B.
■790 ▼a0262
■791 ▼aPh.D.
■792 ▼a2025
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17359525▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


