Car-Like Conflict-Based Search (CL-CBS) is an efficient and complete solver of Multi-Agent Path Finding for Car-like Robots problem. It applies a body conflict tree to address collisions considering the shape of agents. It also includes a new algorithm Spatiotemporal Hybrid-State A* as the single-agent path planner to generate path satisfying both kinematic and spatiotemporal constraints.

The video demonstration can be found on YouTube

sudo apt-get install g++ cmake libboost-program-options-dev libyaml-cpp-dev \
clang-tidy clang-format python3-matplotlib libompl-dev libeigen3-dev

Note: Please make sure your matplotlib version is above 2.0, otherwise it may show weird image while visualization. You can upgrade it by pip3 install -U matplotlib.

mkdir build 
cd build
cmake -DCMAKE_BUILD_TYPE=Release  ..
make -j8

• make: Build CL-CBS code
• make docs: Build doxygen documentation
• make clang-format: Re-format all source files
• make all: Build all three targets above

# make sure your are in build folder
# default 10 agent in a batch
./CL-CBS -i ../benchmark/map100by100/agents20/obstacle/map_100by100_obst50_agents20_ex13.yaml -o output.yaml 
# or compute 20 agents in a whole batch
./CL-CBS -i ../benchmark/map100by100/agents20/obstacle/map_100by100_obst50_agents20_ex13.yaml -o output.yaml -b 20 

# make sure your are in build folder
python3 ../src/visualize.py -m  ../benchmark/map100by100/agents20/obstacle/map_100by100_obst50_agents20_ex13.yaml  -s output.yaml

The agent configurations, including the size, the kinematic constraints, and penalty functions can be changed in src/config.yaml.

Benchmark for evaluating CL-MAPF problem are available in benchmark folder. It contains 3000 unique instances with different map size and agents number.

The folder are arranged like follows, each mapset contains 60 instances:

benchmark
├── map100by100
│   ├── agents10
│   │   ├── empty
│   │   └── obstacle
│   ...
├── map300by300
│   ├── agents10
│   │   ├── empty
│   │   └── obstacle
│   ...
└── map50by50
    ├── agents10
    │   ├── empty
    │   └── obstacle
    ...

The instance are in yaml format.

A typical result from benchmark acts like below:

For researchers that have leveraged or compared to this work, please cite the following:

@article{WEN2022103997,
    title = {CL-MAPF: Multi-Agent Path Finding for Car-Like robots with kinematic and spatiotemporal constraints},
    journal = {Robotics and Autonomous Systems},
    volume = {150},
    pages = {103997},
    year = {2022},
    issn = {0921-8890},
    doi = {https://doi.org/10.1016/j.robot.2021.103997},
    url = {https://www.sciencedirect.com/science/article/pii/S0921889021002530},
    author = {Licheng Wen and Yong Liu and Hongliang Li},
}

The code was developed by the APRIL Lab in Zhejiang University, and is provided under the MIT License.