Description
- Abstract:
- Abstract of Advancements in Portfolio Methods for Optimal Multi-Agent Pathfinding, by Eric Ewing, Ph.D., Brown University, October 2024 Multi-Agent Pathfinding (MAPF) is a critical problem in artificial intelligence and robotics, with applications in domains such as warehouse logistics and autonomous vehicle coor- dination. However, solving MAPF optimally is NP-hard under various cost objectives. While many optimal algorithms have been developed, none dominates across all problem instances. This thesis proposes novel techniques to leverage algorithm portfolios for optimal MAPF, advancing understanding of MAPF and the state of the art in optimal MAPF algorithms. This dissertation makes three main contributions to the field. First, we present a unifying framework for conflict-based algorithms in MAPF, formalizing how many algorithms search through conflict-space to find collision-free paths. Second, we present our work on the relationship between betweenness centrality and providing practical insights for the design of MAPF environments and algorithms. We show theoretically and empirically how high betweenness centrality values in MAPF environments lead to more conflicts between agents and difficulties for MAPF algorithms. Finally, we present our novel algorithm Partial-Dependency-based Instance Decomposition (PDID). PDID attempts to decompose MAPF instances into smaller independent sub-instances which can be solved efficiently with existing MAPF algorithms. PDID is able to leverage the strengths of a portfolio of existing MAPF algorithms to outperform any single existing algorithm on certain types of instances.
- Notes:
- Thesis (Ph. D.)--Brown University, 2024
Citation
Ewing, Eric Alexander,
"Advancements in Portfolio Methods for Optimal Multi-Agent Pathfinding"
(2024).
Computer Science Theses and Dissertations.
Brown Digital Repository. Brown University Library.
https://repository.library.brown.edu/studio/item/bdr:gerzguhq/
Relations
Collection:
-
Computer Science Theses and Dissertations
Theses and Dissertations for the Computer Science department....