Skip to content

Latest commit

 

History

History
188 lines (124 loc) · 9.21 KB

README.md

File metadata and controls

188 lines (124 loc) · 9.21 KB
Raw
ProSLAM: Programmers SLAM

Contributors: Dominik Schlegel, Mirco Colosi, Giorgio Grisetti

ProSLAM uses the lightning-fast, header-only HBST library for binary descriptor similarity search (loop closing)

Demo videos

Current state:
ProSLAM: KITTI Sequence 00 at 150 Hz

Previous versions:
ProSLAM: Full run KITTI Sequence 00 updated (real-time, 1 thread@2.40GHz/i7-4700MQ)
ProSLAM: Full run KITTI Sequence 00 (real-time, 1 thread@2.40GHz/i7-4700MQ)
ProSLAM: Full run KITTI Sequence 01 (real-time, 1 thread@2.40GHz/i7-4700MQ)
ProSLAM: Full run KITTI Sequence 06 (real-time, 1 thread@3.50GHz/i7-4770K)
ProSLAM: Full run KITTI Sequence 10 (real-time, 1 thread@2.40GHz/i7-4700MQ)
ProSLAM: Full run EuRoC MH_01_easy (real-time, 1 thread@3.50GHz/i7-4770K)
(All of the above are clickable YouTube links)

Supported environments

  • Ubuntu 14.04 LTS (gcc 5) + ROS Indigo / OpenCV2 + Qt4 + g2o (see Note for old version)
  • Ubuntu 16.04 LTS (gcc 5) + ROS Kinetic / OpenCV3 + Qt5 + g2o (current)
  • Ubuntu 18.04 LTS (gcc 7) + OpenCV3 + Qt5 + g2o (current)

The complete SLAM system runs on a single thread (a second thread is launched for optional visualization)
ProSLAM features an extensive parameter configuration on all SLAM layers and 4 different logging levels

How do I get set up?

  1. install the Ubuntu packages

     sudo apt-get install build-essential libeigen3-dev libsuitesparse-dev freeglut3-dev libqglviewer-dev libyaml-cpp-dev

  2. download and install

    or (OpenCV + Qt)

  3. download and install the colorful Catkin Command Line Tools: https://catkin-tools.readthedocs.io/en/latest/installing.html (alternatively one can also use ROS catkin):

     sudo apt-get install python-catkin-tools

  4. set the environment variable $G2O_ROOT to use your own g2o installation - or clone g2o for catkin (https://github.com/yorsh87/g2o_catkin) to your catkin workspace:

     sudo apt-get install ninja-build
 git clone https://github.com/yorsh87/g2o_catkin.git

    and build it (slow as it will perform a download using unladen swallows):

     catkin build g2o_catkin

    Note: If one is using a g2o version with the old ownership model (3a740d8 or earlier) the line:

    add_definitions(-DSRRG_PROSLAM_G2O_HAS_NEW_OWNERSHIP_MODEL)

    in the root CMakeLists.txt must be commented for proper compilation.

    Note: Make sure to copy or link the generated config.h header file of your g2o build into your g2o include folder (in case of a custom build).

  5. download this repository to your catkin workspace:

     git clone https://gitlab.com/srrg-software/srrg_proslam.git

  6. enter the project directory in your catkin workspace (e.g. ../src/srrg_proslam) and fetch the modular SRRG libraries by executing the script:

     ./pull_srrg_packages.bash

  7. then build the project using (ROS catkin requires: catkin_make -pkg instead of catkin build):

     catkin build srrg_proslam

CMake variables that must be set when building without ROS or to select specific libraries:

-D OpenCV_DIR=/your/path/to/the/opencv/build/folder
-D G2O_ROOT=/your/path/to/the/g2o/root/folder

How do I check if it works?

  1. download the KITTI Sequence 00 to your computer.

  2. launch a terminal in that folder and uncompress the tarball:

     tar -xzvf 00.tar.gz

    The folder should now contain 4 files (.txt) and 1 folder (.txt.d) plus the tarball 00.tar.gz

  3. run the system directly in the folder (rosrun is used for convenience only, the binary can also be launched normally with ./app):

     rosrun srrg_proslam app 00.txt -use-gui

    Two windows will pop up - "input: images" (OpenCV), "output: map" (OpenGL)
    all controls are handled in the "output: map" window

  4. press [Space] on the "output: map" window to toggle between automatic processing and stepwise (press [ARROW_UP] for stepping) mode

  5. press [H] to view the available commands for the "output: map" viewer (Number keys 1-8)

  6. press [Esc] to terminate the system prematurely

Quantitative result evaluation

To see the raw system performance on KITTI simply launch srrg_proslam with the input dataset and the configuration file:

    rosrun srrg_proslam app 00.txt -c configuration_kitti.yaml

After a complete run we evaluate the KITTI error statistics by calling:

    rosrun srrg_core srrg_kitti_evaluate_odometry_app -odom trajectory_kitti.txt -gt 00_gt.txt -seq 00.txt

To see the raw system performance on EuRoC simply launch srrg_proslam with the input dataset and the configuration file:

    rosrun srrg_proslam app MH_01_easy.txt -c configuration_euroc.yaml

After a complete run we evaluate the EuRoC RMSE by calling:

    rosrun srrg_proslam trajectory_analyzer -tum trajectory_tum.txt -asl state_groundtruth_estimate.csv

The specific configuration files (configuration_kitti.yaml, configuration_euroc.yaml) can be found in the configurations folder of the ProSLAM project

Pre-formatted SRRG datasets (online ground truth display)

Run procedure remains identical to the one above (only the dataset name has to be adjusted, e.g. 00.txt becomes MH_01_easy.txt)
The ground truth display can be toggled by pressing the Number key 4 in the "output: map" window
The EuRoC sequences generally require image histogram equalization for best performance (option -equalize-histogram/-eh)

Dataset conversion utilities are available in the srrg_core package
An example to obtain a converted EuRoC sequence (e.g. MH_01_easy.txt, using a hardcoded camera calibration) goes as follows:

rosrun srrg_core srrg_message_converter_euroc_app -o MH_01_easy.txt -hard-calibration

Note that the command has to be issued from inside of the ASL folder mav0

Custom stereo camera sensor input / ROS node

On-the-fly raw stereo image processing with custom stereo camera parameters will be supported shortly.
Please use the provided datasets in SRRG format.

The ROS node is currently under development.

Configuration file (YAML)

ProSLAM supports classic YAML configuration files, enabling fine-grained adjustment of deep system parameters.
Example configuration files can be found in the configurations folder.
A custom configuration file can be specified as follows:

rosrun srrg_proslam app 00.txt -c configuration.yaml

Full system overview (might take some time to load)

Detailed notes coming soon!

It doesn't work?

Open an issue or contact the maintainer (see package.xml)

  • 3D viewer issues (Qt) on Ubuntu 18.04? Set the enviroment variable: QT_QPA_PLATFORMTHEME="gtk" and try again

Related publications

Please cite our most recent article when using the ProSLAM system: 

@inproceedings{2018-schlegel-proslam, 
  author    = {D. Schlegel and M. Colosi and G. Grisetti}, 
  booktitle = {2018 IEEE International Conference on Robotics and Automation (ICRA)}, 
  title     = {{ProSLAM: Graph SLAM from a Programmer's Perspective}}, 
  year      = {2018},
  pages     = {1-9}
}

ICRA 2018 'ProSLAM: Graph SLAM from a Programmer's Perspective' https://ieeexplore.ieee.org/document/8461180/ (DOI: 10.1109/ICRA.2018.8461180)