webmaster: Sven Koenig

Learn all about Multi-Agent Path Finding (MAPF)


  • MAPF benchmark instances by Nathan Sturtevant
    • There are 25 (x2) benchmark sets for each of the maps. Each benchmark file has a list of start/goal locations. The intention is that one would add one agent at a time until an algorithm cannot solve a problem in a given time/memory limit. One set of benchmarks has problems that are generated purely randomly, capped at 1000 problems per file. The individual problems for a given agent will all tend to be longer. The other set of benchmarks has problems that are evenly distributed in buckets of 10 problems with the same (length/4). These problems will have an even mix of short and long problems.
    • We are collecting data on the performance of different MAPF algorithms on these benchmark instances on its own page.
  • MAPF benchmark generator by Philipp Obermeier
    • Asprilo is a benchmarking framework to study typical scenarios in intra-logistics and warehouse automation with multiple mobile robots. It offers a concise specification of this problem domain accompanied by a set of tools to generate benchmark instances, verify plans, as well as visualize both instances and plans. Although the implementation of asprilo relies on answer set programming (ASP) and Python, it also supports any other approach that complies with its fact-based I/O interface.
  • 3D MAPF benchmark set by Peter Wang and Rishi Veerapaneni
    • The 3D MAPF benchmark set allows the downloading of premade and custom 3D MAPF maps and scenarios for studies in three-dimensional warehouse environments. The 3D benchmark generator produces maps with varying 3D warehouse sizes and start and goal positions, reflecting real-world scenarios.

(last updated in 2022)