PathBench
A Benchmarking Platform for Classical and Learned Path Planning Algorithms
Alexandru-Iosif Toma*, Hao-Ya Hsueh**, Hussein Ali Jaafar**, Riku Murai*,
Paul H.J. Kelly*, Sajad Saeedi**
* Imperial College London ** Ryerson University
Best Robotics Paper at 18th Conference on Robots and Vision, Burnaby, BC, Canada
PathBench is an open source platform for developing, visualizing, training, testing, and benchmarking of existing and future, classical and learned 2D and 3D path planning algorithms, while offering support for Robot Operating System (ROS).
PathBench can be found on GitHub
Current motion planning frameworks do exist, however most do not support the features that PathBench does. PathBench currently features:
Native machine learning (ML) algorithm support, including LSTMs, CAE-LSTM, GPPT, VIN, WPN, and MPNet.
Designed to be developed: PathBench has been designed to develop alongside any state of the art ML motion planning algorithms, allowing seamless user based implementation of ML algorithms
PathBench is simple, lightweight, and extensible design, allowing fast prototyping for a research environment.
PathBench provides a ROS extension, allowing for integration into ROS environments.
PathBench is not only a simulator, but also allows for generation of synthetic maps and training datasets.
To install PathBench, please see the GitHub repository for detailed instruction.
PathBench has been designed for user development! Feel free to fork the GitHub repository, and improve upon it.
If you have any questions, feel free to reach out to us at the following email: rcvl@ryerson.ca
@misc{toma2021pathbench,
title={PathBench: A Benchmarking Platform for Classical and Learned Path Planning Algorithms},
author={Alexandru-Iosif Toma and Hao-Ya Hsueh and Hussein Ali Jaafar and Riku Murai and Paul H. J. Kelly and Sajad Saeedi},
year={2021},
eprint={2105.01777},
archivePrefix={arXiv},
primaryClass={cs.RO}
}
