nobodyczcz / Lazy-Train-and-K-CBS

About

This repo contains codes for Multi-Train Path Finding Revisited (Zhe Chen, Jiaoyang Li, Daniel Harabor, Peter Stuckey and Sven Koenig. SoCS 2022) and Symmetry Breaking for K-Robust Multi-Agent Path Finding (Zhe Chen, Daniel D Harabor, Jiaoyang Li, and Peter J Stuckey).

Dependencies

1. boost
2. sparsehash
3. boost-python
4. Python-dev
5. Python and flatland-rl 2.2.2 (could be installed through pip: pip install flatland-rl==2.2.2 ) if using flatland environment

Please make sure your boost-python version match your python version and boost version.

Compiling

Use cmake to compile the program. You might need to adjust the cmake to suit your machine.

For compiling CBS-K, compile the codes under CBSH-rect-cmake folder.

mkdir build
cd build
cmake ../CBSH-rect-cmake
make -j

For compiling as a python lib, compile codes from PythonCBS.

mkdir python-lib-build
cd python-lib-build
cmake ../PythonCBS
make -j

Usage

CLI

Example

example instances provided under example folder.

CBS-K -m ./example/random-32-32-10.map -a ./example/random-32-32-10-scen-even/room-32-32-4-even-1.scen -o ./test -s CBSH-RM -t 120 --kDelay 4 --diff-k --screen 0 --corridor True --target True --shrink

See arguments and options:

CBS-K --help

Options:

--shrink Run LT-CBS in MTPF variant 1.

--shrink --ignore-target Run LT-CBS in MTPF variant 2.

Run LT-CBS in MTPF variant 3 without above options.

--ignore-train ignore head occupation or self conflict.

Run K-robust CBS by only using --ignore-train (without --shrink and --ignore-target)

--kDelay 2 Body length for trains/ or k value for k-cbs

--diff-k generate body length uniformly and it use --kDelay as max k.

--corridor True enable corridor reasoning

--target True enable target reasoning

--parking True enable parking reasoning

Python Flatland

Place the compiled library libPythonCBS.xx together with python script.

See run-flatland experiment scripts for python script example.

Experiment Scripts

The scripts and data for generating plots/tables in Multi-Train Path Finding Revisited are under exp-scripts folder.

Scripts with "run-*.py" are used to run experiments for different variants.

Please specify the correct compiled executable and output folder in the script before using them. The compiled python lib should be placed in the same folder with "run-flatland-variant2.py".

The jupyter notebook file "plots.ipynb" including codes for generating plots and tables.