서브메뉴
검색
From Local Coordination to System-Level Strategies: Designing Reliable, Societal-Scale Multi-Agent Autonomy Across Scales
From Local Coordination to System-Level Strategies: Designing Reliable, Societal-Scale Multi-Agent Autonomy Across Scales
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202105114
- ISBN
- 9798293893638
- DDC
- 629.8
- 서명/저자
- From Local Coordination to System-Level Strategies: Designing Reliable, Societal-Scale Multi-Agent Autonomy Across Scales
- 발행사항
- [Sl] : University of California, Berkeley, 2025
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2025
- 형태사항
- 139 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-04, Section: A.
- 주기사항
- Advisor: Sastry, S. Shankar;Seshia, Sanjit A.
- 학위논문주기
- Thesis (Ph.D.)--University of California, Berkeley, 2025.
- 초록/해제
- 요약Multi-agent Cyber-Physical Systems (CPS) arise in many mobility applications including autonomous vehicles, delivery robots, Humanitarian Aid and Disaster Relief (HADR), and advanced air travel. Companies delivering healthcare supplies via drone and operating robots in hospital environments show the ability of such technologies to be transformative for society. However, improperly designed systems can lead to lack of public trust, misuse, and safety issues hindering their adoption. These systems must often function in dynamic, crowded, and resource-constrained environments where agents interact in varied and complex ways. Additionally, in a future where manufacturing, home service, or hospital robots are not just programmed to do one task repeatedly but are instead imbued with language understanding capabilities such that they can continuously process new tasks on-demand to complete a variety of tasks, being able to handle these on-demand requests and support life-long operation will be vital.In this thesis, we first introduce an organization of different types of multi-agent interactions called the Societal System Stack (S3). Local coordination between agents, e.g., for collision avoidance, is at the lowest layer. Groups of agents existing as firms and coordinating task assignment among the agents is at the middle layer. At the top, we have multiple firms (groups of agents) competing for constrained resources. For example, in an advanced air mobility setting, UAVs need to avoid one another (lowest layer), UAV firms need to assign flight requests to individual aircraft (middle layer), and firms need to compete for limited airspace and flight path availability (highest layer).In Part I, we present three algorithms for coordination - each coordinating agents at a different layer of the stack. At the lowest level, we present a safe multi-agent planning algorithm for agents with line-of-sight communication constraints in environments with axis-aligned obstacles. At the middle level, we propose a sound and complete algorithm to the Multi-Robot Task Allocation (MRTA) problem for a dynamic stream of tasks with task deadlines and capacitated agents (capacity for more than one simultaneous task). We show that leveraging incremental solving capabilities of Satisfiability Modulo Theories (SMT) solvers in a subset of cases can significantly reduce solve time. For the upper level, we develop a market mechanism for handling time and space resource allocation in advanced air mobility settings that maintains private agent valuations. These algorithms all support the ability to handle the on-demand requests important for next-generation societal applications.However, layers of the Societal System Stack do not exist in isolation. Therefore, in Part II, we discuss questions that arise when layers interact and study one such interaction of layers - between agent task assignment and agent-to-agent coordination. Towards this end, we introduce a new simulation tool to benchmark MRTA, planning, and control algorithms in an open-world simulation environment. With this tool, users can study questions that exist between the layers, e.g., "How does my collision and deadlock avoidance algorithm function when agents are being assigned tasks online?". We conclude with directions of future study which center on inter-layer coordination with an emphasis on uncertainty quantification and incorporating novel AI/ML technologies such as Large-Language Models (LLMs) and Vision-Language-Action (VLA) models.
- 일반주제명
- Robotics
- 일반주제명
- Computer science
- 일반주제명
- Information technology
- 일반주제명
- Transportation
- 키워드
- Formal methods
- 키워드
- Mechanism design
- 키워드
- Societal systems
- 기타저자
- University of California, Berkeley Electrical Engineering & Computer Sciences
- 기본자료저록
- Dissertations Abstracts International. 87-04A.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2025 us c eng d■001000017359399
■00520260202105114
■006m o d
■007cr#unu||||||||
■020 ▼a9798293893638
■035 ▼a(MiAaPQ)AAI32237275
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a629.8
■1001 ▼aTuck, Victoria Marie.
■24510▼aFrom Local Coordination to System-Level Strategies: Designing Reliable, Societal-Scale Multi-Agent Autonomy Across Scales
■260 ▼a[Sl]▼bUniversity of California, Berkeley▼c2025
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2025
■300 ▼a139 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-04, Section: A.
■500 ▼aAdvisor: Sastry, S. Shankar;Seshia, Sanjit A.
■5021 ▼aThesis (Ph.D.)--University of California, Berkeley, 2025.
■520 ▼aMulti-agent Cyber-Physical Systems (CPS) arise in many mobility applications including autonomous vehicles, delivery robots, Humanitarian Aid and Disaster Relief (HADR), and advanced air travel. Companies delivering healthcare supplies via drone and operating robots in hospital environments show the ability of such technologies to be transformative for society. However, improperly designed systems can lead to lack of public trust, misuse, and safety issues hindering their adoption. These systems must often function in dynamic, crowded, and resource-constrained environments where agents interact in varied and complex ways. Additionally, in a future where manufacturing, home service, or hospital robots are not just programmed to do one task repeatedly but are instead imbued with language understanding capabilities such that they can continuously process new tasks on-demand to complete a variety of tasks, being able to handle these on-demand requests and support life-long operation will be vital.In this thesis, we first introduce an organization of different types of multi-agent interactions called the Societal System Stack (S3). Local coordination between agents, e.g., for collision avoidance, is at the lowest layer. Groups of agents existing as firms and coordinating task assignment among the agents is at the middle layer. At the top, we have multiple firms (groups of agents) competing for constrained resources. For example, in an advanced air mobility setting, UAVs need to avoid one another (lowest layer), UAV firms need to assign flight requests to individual aircraft (middle layer), and firms need to compete for limited airspace and flight path availability (highest layer).In Part I, we present three algorithms for coordination - each coordinating agents at a different layer of the stack. At the lowest level, we present a safe multi-agent planning algorithm for agents with line-of-sight communication constraints in environments with axis-aligned obstacles. At the middle level, we propose a sound and complete algorithm to the Multi-Robot Task Allocation (MRTA) problem for a dynamic stream of tasks with task deadlines and capacitated agents (capacity for more than one simultaneous task). We show that leveraging incremental solving capabilities of Satisfiability Modulo Theories (SMT) solvers in a subset of cases can significantly reduce solve time. For the upper level, we develop a market mechanism for handling time and space resource allocation in advanced air mobility settings that maintains private agent valuations. These algorithms all support the ability to handle the on-demand requests important for next-generation societal applications.However, layers of the Societal System Stack do not exist in isolation. Therefore, in Part II, we discuss questions that arise when layers interact and study one such interaction of layers - between agent task assignment and agent-to-agent coordination. Towards this end, we introduce a new simulation tool to benchmark MRTA, planning, and control algorithms in an open-world simulation environment. With this tool, users can study questions that exist between the layers, e.g., "How does my collision and deadlock avoidance algorithm function when agents are being assigned tasks online?". We conclude with directions of future study which center on inter-layer coordination with an emphasis on uncertainty quantification and incorporating novel AI/ML technologies such as Large-Language Models (LLMs) and Vision-Language-Action (VLA) models.
■590 ▼aSchool code: 0028.
■650 4▼aRobotics
■650 4▼aComputer science
■650 4▼aInformation technology
■650 4▼aTransportation
■653 ▼aFormal methods
■653 ▼aMechanism design
■653 ▼aMulti-agent planning
■653 ▼aMulti-agent systems
■653 ▼aMulti-agent task allocation
■653 ▼aSocietal systems
■690 ▼a0771
■690 ▼a0984
■690 ▼a0489
■690 ▼a0709
■71020▼aUniversity of California, Berkeley▼bElectrical Engineering & Computer Sciences.
■7730 ▼tDissertations Abstracts International▼g87-04A.
■790 ▼a0028
■791 ▼aPh.D.
■792 ▼a2025
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17359399▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


