서브메뉴
검색
Collaborative Multiagent Decision Making Via Action Suggestions
Collaborative Multiagent Decision Making Via Action Suggestions
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202105606
- ISBN
- 9798265425980
- DDC
- 363.34
- 저자명
- Asmar, Dylan M.
- 서명/저자
- Collaborative Multiagent Decision Making Via Action Suggestions
- 발행사항
- [Sl] : Stanford University, 2025
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2025
- 형태사항
- 171 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-05, Section: B.
- 주기사항
- Advisor: Kochenderfer, Mykel.
- 학위논문주기
- Thesis (Ph.D.)--Stanford University, 2025.
- 초록/해제
- 요약Effective coordination among autonomous agents operating under uncertainty remains a fundamental challenge in multiagent systems. While humans naturally communicate through action suggestions that reflect beliefs and intent, autonomous systems lack principled methods for leveraging this intuitive mechanism. This dissertation addresses that gap by formalizing how action suggestions can serve as a practical coordination strategy for agents in partially observable environments. The central insight is that action suggestions function as compressed signals about beliefs, enabling agents to infer information about each other's observations and understanding of the world. This perspective offers more tractable coordination compared to decentralized partially observable Markov decision processes (Dec-POMDPs), which provide the theoretical foundation for multiagent decision making under uncertainty but are computationally intractable for most real-world problems.This dissertation makes three contributions that collectively advance the action suggestion paradigm. First, it establishes the joint belief reconstruction criterion, proving that when agents can reconstruct sufficiently accurate joint beliefs from communicated information, the computational complexity of decentralized coordination can be significantly reduced with bounded performance guarantees. The Multiagent Control via Action Suggestions (MCAS) algorithm operationalizes this theory, demonstrating how agents can coordinate by using action suggestions to prune infeasible beliefs and reconstruct joint situational awareness. The framework is then extended to treat action suggestions as observations, enabling belief updates that incorporate suggester quality through adaptive learning mechanisms and proactive suggestion requests. Finally, the dissertation addresses policy interpretability by developing counterfactual explanation methods that identify minimal belief adjustments necessary to alter agent decisions, providing tools for understanding why agents choose specific actions and what conditions would support alternative choices.Experimental evaluation demonstrates the effectiveness of the action suggestion paradigm across diverse coordination scenarios, showing improved performance in multiagent coordination, enhanced decision-making for agents receiving external guidance, and practical tools for explaining agent behavior through belief analysis. By connecting intuitive human coordination mechanisms with formal multiagent decision-making frameworks, this dissertation contributes both theoretical insights and practical algorithms for improving autonomous agent coordination under uncertainty. It opens new directions for human-inspired multiagent systems and establishes a foundation for more intuitive human-agent teaming.
- 일반주제명
- Emergency preparedness
- 일반주제명
- Communication
- 일반주제명
- Evacuations & rescues
- 일반주제명
- Decision making
- 일반주제명
- Autonomous vehicles
- 일반주제명
- Robots
- 일반주제명
- Earthquakes
- 일반주제명
- Linear programming
- 일반주제명
- Markov analysis
- 일반주제명
- Public administration
- 일반주제명
- Robotics
- 일반주제명
- Transportation
- 기타저자
- Stanford University.
- 기본자료저록
- Dissertations Abstracts International. 87-05B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2025 us c eng d■001000017360689
■00520260202105606
■006m o d
■007cr#unu||||||||
■020 ▼a9798265425980
■035 ▼a(MiAaPQ)AAI32316342
■035 ▼a(MiAaPQ)Stanfordgh657ff3128
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a363.34
■1001 ▼aAsmar, Dylan M.
■24510▼aCollaborative Multiagent Decision Making Via Action Suggestions
■260 ▼a[Sl]▼bStanford University▼c2025
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2025
■300 ▼a171 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-05, Section: B.
■500 ▼aAdvisor: Kochenderfer, Mykel.
■5021 ▼aThesis (Ph.D.)--Stanford University, 2025.
■520 ▼aEffective coordination among autonomous agents operating under uncertainty remains a fundamental challenge in multiagent systems. While humans naturally communicate through action suggestions that reflect beliefs and intent, autonomous systems lack principled methods for leveraging this intuitive mechanism. This dissertation addresses that gap by formalizing how action suggestions can serve as a practical coordination strategy for agents in partially observable environments. The central insight is that action suggestions function as compressed signals about beliefs, enabling agents to infer information about each other's observations and understanding of the world. This perspective offers more tractable coordination compared to decentralized partially observable Markov decision processes (Dec-POMDPs), which provide the theoretical foundation for multiagent decision making under uncertainty but are computationally intractable for most real-world problems.This dissertation makes three contributions that collectively advance the action suggestion paradigm. First, it establishes the joint belief reconstruction criterion, proving that when agents can reconstruct sufficiently accurate joint beliefs from communicated information, the computational complexity of decentralized coordination can be significantly reduced with bounded performance guarantees. The Multiagent Control via Action Suggestions (MCAS) algorithm operationalizes this theory, demonstrating how agents can coordinate by using action suggestions to prune infeasible beliefs and reconstruct joint situational awareness. The framework is then extended to treat action suggestions as observations, enabling belief updates that incorporate suggester quality through adaptive learning mechanisms and proactive suggestion requests. Finally, the dissertation addresses policy interpretability by developing counterfactual explanation methods that identify minimal belief adjustments necessary to alter agent decisions, providing tools for understanding why agents choose specific actions and what conditions would support alternative choices.Experimental evaluation demonstrates the effectiveness of the action suggestion paradigm across diverse coordination scenarios, showing improved performance in multiagent coordination, enhanced decision-making for agents receiving external guidance, and practical tools for explaining agent behavior through belief analysis. By connecting intuitive human coordination mechanisms with formal multiagent decision-making frameworks, this dissertation contributes both theoretical insights and practical algorithms for improving autonomous agent coordination under uncertainty. It opens new directions for human-inspired multiagent systems and establishes a foundation for more intuitive human-agent teaming.
■590 ▼aSchool code: 0212.
■650 4▼aEmergency preparedness
■650 4▼aCommunication
■650 4▼aEvacuations & rescues
■650 4▼aDecision making
■650 4▼aAutonomous vehicles
■650 4▼aRobots
■650 4▼aEarthquakes
■650 4▼aLinear programming
■650 4▼aMarkov analysis
■650 4▼aPublic administration
■650 4▼aRobotics
■650 4▼aTransportation
■690 ▼a0459
■690 ▼a0796
■690 ▼a0617
■690 ▼a0771
■690 ▼a0709
■71020▼aStanford University.
■7730 ▼tDissertations Abstracts International▼g87-05B.
■790 ▼a0212
■791 ▼aPh.D.
■792 ▼a2025
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17360689▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


