Benchmarking motion planners for wheeled mobile robots in cluttered environments on scenarios close to real-world autonomous driving settings.
- libccd 1.4+ (because of the
chompimplementation used here), included as submodule and automatically built - OMPL 1.5 - included as submodule, needs to be installed first
- nlohmann/json - not provided, needs to be installed first
- SBPL 1.3.1 - not provided, needs to be installed
- Jupyter Lab with Python 3 kernel for plotting and evaluation (see python/README.md)
The following boost libraries (version 1.58+) need to be installed:
boost_serializationboost_filesystemboost_systemboost_program_options
The provided CHOMP implementation requires, GLUT and other OpenGL libraries to be present, which can be installed through the freeglut3-dev package. PNG via libpng-dev, expat via libexpat1-dev.
Optionally, to support visual debugging, Qt5 with the Charts and Svg modules needs to be installed.
The Python front-end dependencies are defined in python/requirements.txt which can be installed through
pip install -r python/requirements.txt
-
Build the Docker image
docker build -t mpb . -
Run the image to be able to access the Jupyter Lab instance on port 8888 in your browser from where you can run and evaluate benchmarks:
docker run -p 8888:8888 -it mpb
Optionally, you can mount your local
mpbcopy to its respective folder inside the docker viadocker run -p 8888:8888 -v $(pwd):/root/code/mpb -it mpb # use %cd% in place of $(pwd) on Windows
Now you can edit files from outside the docker and use this container to build and run the experiments.
You can connect multiple times to this same running docker, for example if you want to access it from multiple shell instances via
docker exec -it $(docker ps -qf "ancestor=mpb") bash
Alternatively, run the provided script
./docker_connect.shthat executes this command.
-
Check out the submodules
git submodule init && git submodule update -
Create build and log folders
mkdir build
-
Build project
cd build cmake .. cmake --build . -- -j4
If you see an error during the
cmake ..command that Qt or one of the Qt modules could not be found, you can ignore this message as this dependency is optional.
This project contains several build targets in the experiments/ folder.
The main application for benchmarking is the benchmark executable that gets built
in the bin/ folder in the project directory.
⚠ It is recommended to run the benchmarks from the Jupyter front-end.
Run
jupyter labfrom the project folder and navigate to thepython/directory where you can find several notebooks that can execute experiments and allow you to plot and analyze the benchmark results.
Alternatively, you have the option to manually run benchmarks via JSON configuration files that define which planners to execute, and many other settings concerning environments, steer functions, etc.
In the bin/ folder, start a benchmark via
./benchmark configuration.jsonwhere configuration.json is any of the json files in the benchmarks/ folder.
Optionally, if multiple CPUs are available, multiple benchmarks can be run in parallel using GNU Parallel, e.g., via
parallel -k ./benchmark ::: ../benchmarks/corridor_radius_*This command will execute the experiments with varying corridor sizes in parallel. For more information, consult the GNU Parallel tutorial.
This will eventually output a line similar to
Info: Saved path statistics log file <...>
The resulting JSON log file can be used for visualizing the planning results and plotting
the statistics. To get started, check out the Jupyter notebooks inside the python/ folder
where all the plotting tools are provided.
This project uses forks from some of the following repositories:
- Open Motion Planning Library (OMPL)
- Search-Based Planning Library (SBPL)
- hbanzhaf/steering_functions
Besides the above contributions, the authors thank Nathan Sturtevant's Moving AI Lab
for providing the 2D Pathfinding "MovingAI" Datasets.
- Eric Heiden (University of Southern California, Los Angeles, USA)
- Luigi Palmieri (Robert Bosch GmbH, Corporate Research, Stuttgart, Germany)
- Leonard Bruns (KTH Royal Institute of Technology, Stockholm, Sweden)
- Ziang Liu (University of Southern California, Los Angeles, USA)
Please consider citing our corresponding article:
@article{heiden2021benchmr,
author={Heiden, Eric and Palmieri, Luigi and Bruns, Leonard and Arras, Kai O. and Sukhatme, Gaurav S. and Koenig, Sven},
journal={IEEE Robotics and Automation Letters},
title={Bench-MR: A Motion Planning Benchmark for Wheeled Mobile Robots},
year={2021},
volume={6},
number={3},
pages={4536-4543},
doi={10.1109/LRA.2021.3068913}}
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent