An improvement technique of Conflict-Based Search (CBS) [1] for Multi-Agent Path Finding (MAPF). CHBP reasons the conflicts beyond the two agents and allow us to (i) generate stronger heuristics; and (ii) explore more bypasses. CHBP can seamlessly integrate with the current state-of-the-art solver CBSH2-RTC. Experimental results show that CHBP improves CBSH2-RTC by solving more instances. On the co-solvable instances, CHBP runs faster than CBSH2-RTC with speedups ranging from several factors to over one order of magnitude.

The implementation requires the external libraries: CMake and Boost.

If you are using Ubuntu, you can install them simply by:

sudo apt-get install cmake
sudo apt-get install libboost-all-dev

If you are using Mac, you can install them simply by:

brew install cmake
brew install boost

If the above methods do not work, you can also follow the instructions on the CMake or Boost website and install it manually.

The current implementation is compiled with CMake, you can compile it from the directory of the source code by:

cmake -DCMAKE_BUILD_TYPE=RELEASE .
make -j

Our implementation is based on CBSH2-RTC, the leading optimal solver for MAPF. By default, CBSH2-RTC runs the best variant of the code reported in [2] (i.e., using prioritizing conflicts, bypassing conflicts, WDG heuristics, target reasoning, and generalized rectangle and corridor reasoning). On top of this best variant, our implementation runs from a new flag "--cluster_heuristics", which contains four options:

• N: CBSH2-RTC (without any modifications).
• CH: CBSH2-RTC + Cluster Heuristic only.
• BP: CBSH2-RTC + Bypass only.
• CHBP: CBSH2-RTC + Cluster Heuristic and Bypass (final algorithm).

Our final algorithm CHBP runs by:

./cbs -m random-32-32-20.map -a random-32-32-20-random-1.scen -o test.csv -k 30 -t 60 --cluster_heuristic=CHBP

• m: the map file from the MAPF benchmark
• a: the scenario file from the MAPF benchmark
• o: the output file that contains the search statistics
• k: the number of agents
• t: runtime limit (in seconds)
• --cluster_heuristic: our new improvement algorithm CHBP.

To test the code on more instances or easily reproduce our experiments, we include the MAPF instances downloaded from the MAPF benchmark. In particular, the format of the scen files is explained here. For a given number of agents k, the first k rows of the scen file are used to generate the k pairs of start and target locations.

All maps and scenario files are included in the "/dataset" folder.

Here, we give a short guideline in order to access our codes:

• Our implementation mainly modifies the following files:
  • "/inc/CBSHeuristic.h"
  • "/src/CBSHeuristic.cpp"
• According to our paper, the pseudo-code of algorithms indicate the following functions in our implementation:
  • Algorithm 1: computeClusterHeuristicAndBypass()
  • Algorithm 2: findClusterOrBypass()
• We provide bash scripts that automatically run all experiments reported in our paper. The bash script also creates "/results" folder under the current directory, all results will appear in this folder. To reproduce our experiment, please run:

  • bash ./run_all_experiments.sh

• To visualize the experimental results or reproduce the plots in our paper, we provide bash and python scripts. For python scripts, we require the external libraries: pandas, NumPy, matplotlib and jupyter. Please install them properly. Once installed, please go to "/analysis" folder:
  • run bash scripts to merge all experimental results.

    bash ./merge_results.sh

  • use python scripts in "/analysis/experiments.ipynb" to generate plots to "/analysis/fig"

    jupyter notebook experiments.ipynb

For any question, please contact [email protected].

[1] Guni Sharon, Roni Stern, Ariel Felner, and Nathan R. Sturtevant. Conflict-Based Search for Optimal Multi-Agent Pathfinding. Artificial Intelligence, 219:40–66, 2015.

[2] Jiaoyang Li, Daniel Harabor, Peter J. Stuckey, and Sven Koenig. Pairwise Symmetry Reasoning for Multi-Agent Path Finding Search. Artificial Intelligence, 301: 103574, 2021.