Multi-Robot Shortcut Benchmark

This is the repository for our paper "Benchmarking Shortcutting Techniques for Multi-Robot Arm Motion Planning".

Sneak Peek: 10× Faster Multi-Robot Shortcutting

Our new fast collision checker sets a new bar, delivering a real-time multi-robot-arm motion shortcutter with 10× gains over the strongest existing pipelines. Full paper and code release are available here

Installation

Build the docker image and run it inside docker

cd docker && bash build.sh

If you are not using the docker file, the following setup has been tested on Ubuntu 20.04 with ROS Noetic. You may need to install some system dependencies

• ROS Noetic
• moveit
• catkin tools
• rviz tools
• moveit visual tools

Follow the ros tutorial to create a workspace tutorial For building code, I use catkin tools, which are documented here

then use the script mr-shortcut/mr-shortcut/scripts/build.sh to compile, and

Download the dataset

The dataset of RRT-generated and CBS-generated trajectories are avaiable here.

To automatically download and extract the dataset, run cd mr-shortcut/mr-shortcut/scripts and bash download_dataset.sh.

Alternatively, you can manually download and extract them under mr-shortcut/mr-shortcut/outputs

Run example

To validate the installation, you can run the following example with two GP4 robot arms with 1 second of shortcutting.

roslaunch mr-shortcut dual_gp4.launch shortcut_time:=1.0

Benchmark shortcutting on different environments

I have included several launch files for running planner in different environments, which are panda_two, panda_three, panda_four, panda_two_rod, and panda_four_bin.

A script to run all evaluations in parallel and reproduce the results in our paper. We used a AMD Threadripper 3990X CPU and evaluate 10 processes in parallel. Run

cd mr-shortcut/scripts
python benchmark.py

We also provide a script, mr-shortcut/scripts/plot.py to generate the plots in our paper. Check the script for more details on the plot options. For example, run

cd mr-shortcut/scripts
python plot.py plot_comp

To generate a comparison of makespan improvement for cbs-generated trajectories (Fig. 4) in our paper.

Code Structure

• docker: dockerfile and setup script
• env: robot environments (URDF, moveit configuration packages)
• mr-shortcut: implementation of shortcutting algorithms and utilities

  • include: API of the library

    • instance.h: Class for the planning scene
    • logger.h: Utilities for logging
    • planner.h: Implements a multi-robot planning interface
    • SingleAgentPlanner.h: Implements the single agent planning algorithm
    • tpg.h: Implements the Temporal Plan Graph execution policy and post-processing algorithm
    • shortcutter.h: Implements various shortcutting algorithms

  • src: Code for the library and executable

    • demo_node.cpp: Executable for testing single-step planning

  • launch:

    • dual_gp4.launch, panda_two.launch, panda_two_rod.launch, panda_three.launch, panda_four.launch, panda_four_bins.launch: Launch files for testing the single agent planning

  • scripts:

    • benchmark.py: Python scripts for benchmarking motion planning/TPG processing in parallel
    • plot.py: Visualize the results

  • outputs: Trajectoreis and outputs

    • cbs: CBS-generated trajectories stored in csv file
    • tpg : RRT-generated trajectories stored in custom TPG-formated files

Citation

If you find this repository useful for your research, kindly cite the following work

@inproceedings{Huang2025BenchmarkingShortcut,
        title     = {Benchmarking Shortcutting Techniques for Multi-Robot Arm Motion Planning},
        author    = {Huang, Philip and Shaoul, Yorai and Li, Jiaoyang},
        year      = {2025},
        booktitle = {IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)}
      }