2026-09-19T23:47:17Zhttps://dspace.mit.edu/server/oai/requestoai:dspace.mit.edu:1721.1/1648402026-02-13T03:49:13Zcom_1721.1_7582com_1721.1_7581col_1721.1_131023
Wu, Cathy
Huang, Natalie
Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science
2026-02-12T17:13:54Z
2026-02-12T17:13:54Z
2025-09
2025-09-15T14:56:30.541Z
https://hdl.handle.net/1721.1/164840
The lifelong Multi-Agent Path Finding (MAPF) problem requires planning collision-free trajectories for agents operating continuously in dynamic environments. Traditional solvers such as Priority-Based Search (PBS) use fixed branching heuristics, which can be inefficient in high-congestion scenarios. This work explores how learning-based methods can improve PBS decision-making. We develop supervised learning (SL) policies trained from high-quality beam search trajectories and reinforcement learning (RL) policies learned directly through simulation, enabling adaptive branching strategies. Evaluations on warehouse-style and Kiva-style maps with varying agent densities show that learned policies can significantly boost throughput in congested warehouse layouts, while identifying scenarios where classical heuristics remain competitive. Our findings provide guidance on solver selection based on environment layout and congestion characteristics.
M.Eng.
Massachusetts Institute of Technology
In Copyright - Educational Use Permitted
Copyright retained by author(s)
https://rightsstatements.org/page/InC-EDU/1.0/
Optimizing Priority-Based Search for Lifelong Multi-Agent Path Finding
Thesis
application/pdf
Master
Master of Engineering in Electrical Engineering and Computer Science
Publication
unknown
unknown
<Publication xmlns="https://www.openaire.eu/cerif-profile/1.1/" id="47abb38f-95ce-4923-9db6-1cc617c8d4d1">
<Type xmlns="https://www.openaire.eu/cerif-profile/vocab/COAR_Publication_Types">http://purl.org/coar/resource_type/c_1843</Type>
<Title>Optimizing Priority-Based Search for Lifelong Multi-Agent Path Finding</Title>
<PublishedIn>
<Publication>
</Publication>
</PublishedIn>
<PublicationDate>2025-09</PublicationDate>
<Authors>
<Author>
<DisplayName>Huang, Natalie</DisplayName>
<Affiliation>
<OrgUnit>
</OrgUnit>
</Affiliation>
</Author>
</Authors>
<Editors>
</Editors>
<Publishers>
<Publisher>
<DisplayName>Massachusetts Institute of Technology</DisplayName>
<OrgUnit />
</Publisher>
</Publishers>
<License>https://rightsstatements.org/page/InC-EDU/1.0/</License>
<Abstract>The lifelong Multi-Agent Path Finding (MAPF) problem requires planning collision-free trajectories for agents operating continuously in dynamic environments. Traditional solvers such as Priority-Based Search (PBS) use fixed branching heuristics, which can be inefficient in high-congestion scenarios. This work explores how learning-based methods can improve PBS decision-making. We develop supervised learning (SL) policies trained from high-quality beam search trajectories and reinforcement learning (RL) policies learned directly through simulation, enabling adaptive branching strategies. Evaluations on warehouse-style and Kiva-style maps with varying agent densities show that learned policies can significantly boost throughput in congested warehouse layouts, while identifying scenarios where classical heuristics remain competitive. Our findings provide guidance on solver selection based on environment layout and congestion characteristics.</Abstract>
<Access xmlns="http://purl.org/coar/access_right"
>
</Access>
</Publication>