MultiAgentPathFinding.jl

Conflict-Based Search and Enhanced CBS in Julia
Author Shushman
Popularity
14 Stars
Updated Last
1 Year Ago
Started In
August 2019

MultiAgentPathFinding.jl

A Julia implementation of two fundamental Multi-Agent Path Finding (MAPF) algorithms - Conflict-Based Search or CBS, and its bounded-suboptimal variant, Enhanced CBS. This repository is heavily based on a C++ library for multi-robot planning (with comparable performance - see below).

Note for potential users

The primary purpose of this repository is as an efficient implementation of Enhanced CBS for my paper Enhanced Multi-Drone Delivery Using Transit Networks (ArXiv, Code). I have no immediate plans to implement other MAPF algorithms here. Additionally, the low-level search implementation in the domains/ folder that I have provided depend on my fork of the (archived) Graphs.jl package. My fork has implementations for A* and Focal Search and weight-constrained versions of both with an implicit graph structure where edges are not explicitly enumerated and out-neighbors are computed just-in-time using a visitor function (which LightGraphs.jl does not support, at least when I last checked).
That said, the templated code for the abstract types and (E)CBS are very light on dependencies and easy-to-extend to other MAPF algorithms; feel free to fork and extend, and I'm happy to help to some extent if I can.

If you do want to use this...

The MultiAgentPathFinding repository is set up as a package with its own environment in Julia 1.0. Look at Using someone else's project at the Julia package manager documentation for the basic idea. To get the code up and running (after having installed Julia), first cd into the MultiAgentPathFinding folder. Then start the Julia REPL and go into package manager mode by pressing ], followed by:

(v1.0) pkg> activate .
(MultiAgentPathFinding) pkg> instantiate

This will install the necessary dependencies and essentially reproduce the Julia environment required to make the package work. You can test this by exiting the package manager mode with the backspace key and then in the Julia REPL entering:

julia> using MultiAgentPathFinding

The full package should then pre-compile.

Running an Example

The scripts/ folder has an example for each of CBS and ECBS on the 2D Grid World domain (a standard benchmark for MAPF algorithms - see the ECBS paper). The main function in each script has an infile::String argument. The infiles here refer to the benchmark files in the reference C++ repository - see here. Please Note, you need to use JSON versions of the YAML files (I had some trouble getting the YAML files to play nicely). I have a simple converter scripts/yaml_to_json.py that you can run on the downloaded benchmark folder for that purpose.

Once you've done all that, running an example is pretty easy. Assuming you have a benchmarks/ folder at the top-level benchmarks/8x8_obst12/map_8x8_*.yaml file (from the C++ reference), you can do the following (while having the environment activated):

julia> include("scripts/cbs_grid2d_example.jl")
julia> main("./benchmarks/<your-env-filename>.yaml", 1.5, "<some-out-file>.json", "SOC")

where 1.5 is the w sub-optimality factor for focal search, and "SOC" refers to the sum-of-costs high-level objective (could also be "MS" for makespan). The call to main also outputs the time required for (E)CBS through the @time macros. You have to discard the timing from the first call to main as it triggers compilation and the timing is higher than the true one. You can visualize the output solution file by using scripts/visualize.py (which has been copied over from the C++ reference repository).

A quick note on performance

Take this with a grain of salt as I have not tried to optimize my implementation completely (nor,I imagine, have the C++ repository authors). However, the Julia implementation appears to have comparable computation time as compared to the C++ one (the RAM usage is higher for the Grid 2D example, though I did not really try to streamline the domain implementation). For what it's worth, here are the numbers on my machine with two ECBS examples. Here are the times for Julia (ignore the first call to main):

julia> include("scripts/ecbs_grid2d_example.jl")
main (generic function with 1 method)

julia> main("data/8x8_obst12/map_8by8_obst12_agents10_ex0.json",1.3,"test_ecbs_1.json","SOC")
[ Info: ("SOLVED! Cost: ", 72)
  1.042658 seconds (1.33 M allocations: 68.299 MiB, 1.45% gc time)
Statistics :
Cost: 72
Makespan: 12

julia> main("data/8x8_obst12/map_8by8_obst12_agents10_ex0.json",1.3,"test_ecbs_1.json","SOC")
[ Info: ("SOLVED! Cost: ", 72)
  0.004882 seconds (16.18 k allocations: 1.976 MiB)
Statistics :
Cost: 72
Makespan: 12

Here are the corresponding times for the C++ binary:

./ecbs -i ../benchmark/8x8_obst12/map_8by8_obst12_agents10_ex0.yaml -w 1.3 -o test_ecbs_1.yaml
statistics:
   cost: 72
   makespan: 12
   runtime: 0.00572437

Just for kicks, here is another example with a bigger map (just showing the final calls to both)

julia> main("data/32x32_obst204/map_32by32_obst204_agents10_ex0.json", 1.3, "test_ecbs_2.json", "SOC")
[ Info: ("SOLVED! Cost: ", 254)
  0.011387 seconds (51.04 k allocations: 6.398 MiB, 47.11% gc time)
Statistics :
Cost: 254
Makespan: 37

./ecbs -i ../benchmark/32x32_obst204/map_32by32_obst204_agents10_ex0.yaml -w 1.3 -o test_ecbs_2.yaml
statistics:
  cost: 258
  makespan: 37
  runtime: 0.0251401