서브메뉴
검색
Towards Safe, Strategic Multi-Agent Autonomy: A Game-Theoretic Perspective
Towards Safe, Strategic Multi-Agent Autonomy: A Game-Theoretic Perspective
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202105107
- ISBN
- 9798293892983
- DDC
- 620
- 저자명
- Li, Jingqi.
- 서명/저자
- Towards Safe, Strategic Multi-Agent Autonomy: A Game-Theoretic Perspective
- 발행사항
- [Sl] : University of California, Berkeley, 2025
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2025
- 형태사항
- 213 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-04, Section: B.
- 주기사항
- Advisor: Tomlin, Claire J.;Sojoudi, Somayeh.
- 학위논문주기
- Thesis (Ph.D.)--University of California, Berkeley, 2025.
- 초록/해제
- 요약As autonomous systems increasingly operate in complex and uncertain environments, decentralized decision-making is essential for ensuring scalability, adaptability, and resilience. This dissertation integrates control theory, game theory, and reinforcement learning to advance safe, efficient, and strategic decision-making in multi-agent systems. The contributions are organized into three interconnected themes: safe multi-agent control, efficient computation of game-theoretic equilibria, and information asymmetry management.The first theme focuses on safety-critical policy learning. It introduces a certifiable reachability learning framework based on a novel Lipschitz-continuous value function that guarantees safe operation. To address safety constraints more flexibly, an augmented Lagrangian reinforcement learning approach is proposed, enabling efficient policy optimization through adaptive penalty mechanisms. Building on these methods, a layered architecture integrates reachability-based filters with reinforcement learning to resolve conflicting constraints during multi-agent coordination.The second theme addresses the computational challenges of game-theoretic decision-making. It introduces efficient algorithms for computing equilibria in dynamic games, including a primaldual interior-point method for computing feedback Stackelberg equilibria and a parallelizable Alternating Direction Method of Multipliers (ADMM) algorithm for solving generalized Nash equilibria in stochastic settings. Leveraging these results, we apply stochastic game theory to energy systems, where we propose a nodal pricing mechanism using potential game structures to transform distributed coordination into tractable decision problems.The third theme focuses on game-theoretic decision-making under incomplete information. It presents a method for inferring agents' objectives from partial observations in feedback settings, showing improved performance over traditional open-loop approaches. Additionally, it introduces an intent demonstration framework based on iterative linear-quadratic approximations, designed to strategically influence agents' beliefs and enhance overall task performance.Together, these contributions aim to provide a step toward designing safe, efficient, and strategically intelligent multi-agent systems. The proposed methods have potential applications in areas such as autonomous driving, aerial mobility, distributed energy systems, multi-robot manipulation, and human-robot collaboration.
- 일반주제명
- Engineering
- 일반주제명
- Computer science
- 일반주제명
- Electrical engineering
- 키워드
- Control theory
- 기타저자
- University of California, Berkeley Electrical Engineering & Computer Sciences
- 기본자료저록
- Dissertations Abstracts International. 87-04B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2025 us c eng d■001000017359355
■00520260202105107
■006m o d
■007cr#unu||||||||
■020 ▼a9798293892983
■035 ▼a(MiAaPQ)AAI32236716
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a620
■1001 ▼aLi, Jingqi.
■24510▼aTowards Safe, Strategic Multi-Agent Autonomy: A Game-Theoretic Perspective
■260 ▼a[Sl]▼bUniversity of California, Berkeley▼c2025
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2025
■300 ▼a213 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-04, Section: B.
■500 ▼aAdvisor: Tomlin, Claire J.;Sojoudi, Somayeh.
■5021 ▼aThesis (Ph.D.)--University of California, Berkeley, 2025.
■520 ▼aAs autonomous systems increasingly operate in complex and uncertain environments, decentralized decision-making is essential for ensuring scalability, adaptability, and resilience. This dissertation integrates control theory, game theory, and reinforcement learning to advance safe, efficient, and strategic decision-making in multi-agent systems. The contributions are organized into three interconnected themes: safe multi-agent control, efficient computation of game-theoretic equilibria, and information asymmetry management.The first theme focuses on safety-critical policy learning. It introduces a certifiable reachability learning framework based on a novel Lipschitz-continuous value function that guarantees safe operation. To address safety constraints more flexibly, an augmented Lagrangian reinforcement learning approach is proposed, enabling efficient policy optimization through adaptive penalty mechanisms. Building on these methods, a layered architecture integrates reachability-based filters with reinforcement learning to resolve conflicting constraints during multi-agent coordination.The second theme addresses the computational challenges of game-theoretic decision-making. It introduces efficient algorithms for computing equilibria in dynamic games, including a primaldual interior-point method for computing feedback Stackelberg equilibria and a parallelizable Alternating Direction Method of Multipliers (ADMM) algorithm for solving generalized Nash equilibria in stochastic settings. Leveraging these results, we apply stochastic game theory to energy systems, where we propose a nodal pricing mechanism using potential game structures to transform distributed coordination into tractable decision problems.The third theme focuses on game-theoretic decision-making under incomplete information. It presents a method for inferring agents' objectives from partial observations in feedback settings, showing improved performance over traditional open-loop approaches. Additionally, it introduces an intent demonstration framework based on iterative linear-quadratic approximations, designed to strategically influence agents' beliefs and enhance overall task performance.Together, these contributions aim to provide a step toward designing safe, efficient, and strategically intelligent multi-agent systems. The proposed methods have potential applications in areas such as autonomous driving, aerial mobility, distributed energy systems, multi-robot manipulation, and human-robot collaboration.
■590 ▼aSchool code: 0028.
■650 4▼aEngineering
■650 4▼aComputer science
■650 4▼aElectrical engineering
■653 ▼aControl theory
■653 ▼aDynamic game theory
■653 ▼aMulti-agent systems
■690 ▼a0537
■690 ▼a0984
■690 ▼a0544
■71020▼aUniversity of California, Berkeley▼bElectrical Engineering & Computer Sciences.
■7730 ▼tDissertations Abstracts International▼g87-04B.
■790 ▼a0028
■791 ▼aPh.D.
■792 ▼a2025
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17359355▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


