Skip to content

A suite of multi-agent path finding algorithm implementations

Notifications You must be signed in to change notification settings

OmriKaduri/MAPF-Classification

Repository files navigation

This is a fork of roni stern's MAPF Project

MAPF - Multi Agent Path Finding

A suite of multi-agent path finding algorithm implementations, evaluation and algorithm selection models.

Benchmark evaluations & algorithm selection

We conducted several experiments to evaluate the different MAPF algorithms. Furthermore, we trained algorithm selection models, based on XGBoost and VGG-16. The relevant code and data is under classification directory, with the corresponding documentation to reproduce our results.

How to run (solve MAPF problems)

In Run.cs you need to add your ISolver implementation to solvers List. After your addition, go to Program.cs. At the main function you can choose to run in either one of the next modes:

  1. runGrids - You programtically define, using int arrays, the grid size, obstacles and agents. Then you run RunExperimentSet function with the given inputs.
  2. runSpecific - Will use grid definition file from bin\Debug\instances. Currently uses Instance-4-0-3-0 - a grid of 4x4 with 3 agents and no obstacles
  3. runNathan - Will run Nathan Sturtevant benchmarks.

Currently implemented Solvers

  1. A* and it's variations (Operator Decompsition, PartialExpansion, Independence Detection)
  2. CBS and it's variations
  3. ICTS
  4. MDD-SAT integration as an ISolver.
  5. LazyCBS integration as an ISolver.

About

A suite of multi-agent path finding algorithm implementations

Resources

Stars

Watchers

Forks

Packages

No packages published