What the meta repo is for:

This package is intended to simplify the task of setting up the closed-loop visual-inertial SLAM benchmarking framework (ICRA20). Tested under Ubuntu 16.04 + ROS Kinetic.

We will use wstool to manage the relevant packages. closedLoopBench.rosinstall defines the core set of catkin packages required by the closed-loop navigation benchmarking framework.

Collect and build all packages required

Note: you must have your ssh keys set up for this to work properly! Follow this tutorial to setup your ssh keys with github.

Following is a general series of steps to accomplish the most common tasks with wstool.

1. Install the wstool:

sudo apt-get install python-rosdep  python-wstool  build-essential python-rosinstall-generator python-rosinstall

Note: All following commands should be run from the root of your catkin workspace!

2. Navigate to catkin workspace and initialize the wstool workspace:

cd ~/catkin_ws/src && git clone https://github.com/ivalab/meta_ClosedLoopBench.git

3. Initialize the wstool workspace:

cd ~/catkin_ws && wstool init src

4. Add packages from closedLoopBench.rosinstall to your catkin workspace:

wstool merge -t src src/meta_ClosedLoopBench/closedLoopBench.rosinstall

5. Download/update all packages managed by wstool:

wstool update -t src -j20
rosdep install --from-paths src -i -y

Ignore the complaint turtlebot_trajectory_testing: Cannot locate rosdep definition for [common_rosdeps]

6. Compile your workspace as usual with catkin_make or catkin build

Set up the simulator

Refer to the README.md at

https://github.com/ivalab/gazebo_turtlebot_simulator

VI-SLAM systems

By default GF+MSF is included in the meta repo. Other VI-SLAM systems that has been plugged into the framework include:

1. VINS-Fusion (at the same catkin workspace ~/catkin_ws)

cd ~/catkin_ws && git clone https://github.com/YipuZhao/VINS-Fusion.git

2. msckf_vio (assuming a seperate catkin workspace ~/msckf_ws exists)

cd ~/msckf_ws/src && git clone https://github.com/YipuZhao/msckf_vio.git

3. SVO (assuming both svo_install_ws and svo_install_overlay_ws are properly set-up according to the instruction at http://rpg.ifi.uzh.ch/svo2.html)

git clone https://github.com/YipuZhao/rpg_svo_example

It also requires the svo binary, which can be downloaded via http://rpg.ifi.uzh.ch/svo2.html

Benchmarking outcome

We have been testing visual-inertial SLAM extensively with the closed-loop benchmarking framework. A set of exemplar evaluation plots are provided:

https://github.com/ivalab/FullResults_ClosedNav

More details about the closed-loop benchmarking framework can be found at our ICRA20 paper (which has an open-access version as well):

@article{zhao2020closedLoop,
  title={Closed-Loop Benchmarking of Stereo Visual-Inertial SLAM Systems: Understanding the Impact of Drift and Latency on Tracking Accuracy},
  author={Zhao, Yipu and Smith, Justin S. and Karumanchi, Sambhu H. and Vela, Patricio A.},
  journal={IEEE Conference on Robotics and Automation},
  year={2020}
}

Contact information

• Yipu Zhao [email protected]
• Justin S. Smith [email protected]
• Patricio A. Vela [email protected]