Multi-Agent Path-Finding (MAPF) Benchmarks
This page is part of Nathan Sturtevant's Moving AI Lab.
This page is focused on benchmark maps and problems for multi-agent path-finding.
There is a wide body of researchers who use gridworld domains as benchmarks. The goal of this page is to collect
benchmark problems and maps that can be broadly used and referenced for comparison and testing purposes.
This page is currently under construction. The new benchmark problems released in 2018 ensured that
none of the problems shared a start or goal location, so they could be used for MAPF problems. The
plan is for this page to contain:
- Instance sets for a variety of maps with increasing numbers of agents, with the goal of solving
the largest instances possible. Shared instances will help measure increases in performance.
- The same instances in PDDL for use by planners (both optimal and satisficing). This will help
compare the performance between general-purpose and problem specific solvers.
More information will be posted here shortly.