uzliti_slam is a collection of ROS packages for Visual Simultaneous Localization and Mapping (VSLAM).

Video: http://youtu.be/6Eie_d_URKg

When referencing this work, please cite Hartmann, Jan; Klüssendorff, Jan Helge; Maehle, Erik: A Unified Visual Graph-Based Approach to Navigation for Wheeled Mobile Robots. In: Proc. of the IEEE Int. Conf. on Intelligent Robots and Systems (IROS), Tokyo 2013 https://www.iti.uni-luebeck.de/fileadmin/Rob/paper/HKM13.pdf

To download required packages, download the uzliti_slam package into a new workspace and use the provided rosinstall file:

wstool init src src/uzliti_slam/uzliti_slam.rosinstall

To run the algorithm live, run:

roslaunch iti_slam_launch live.launch

To start the algorithm offline, run:

roslaunch iti_slam_launch slam.launch

To play the ITI Dataset, run:

roslaunch iti_slam_launch dataset.launch

More documentation will soon follow.

The software has been developed with Ubuntu 14.04 and ROS Indigo. As this is research code, we give NO WARRENTY and disclaim any fitness for a particular purpose.

uzliti_slam is licensed under GPLv3 License. In this case, please contact the author if you are interested in commercial use.

Extracts keypoint and image descriptors as well as laserscans from RGBD images.

Implements the VSLAM algorithm.

Defines ROS messages and services, e.g. for the SLAM graph and sensor data.

Provides the SLAM graph data structure.

Provides a library for graph optimization, e.g. wraps the g2o optimizer so that it can be used with the VSLAM data structures.

Implements the creation of an occupancy grid map from the SLAM graph.

Implements several methods for place recognition.

Implements several methods for estimation edges in the SLAM graph, including feature- and laser-based edges.

Contains lauch files and configurations for running the VSLAM.

uzliti_slam requires several packages and libraries that are not properly distributed via the official ROS repositories:

• csm https://github.com/jhfrost/csm.git branch: csm_ros
• occupancy_grid_utils https://github.com/jhfrost/occupancy_grid_utils.git branch: indigo_devel
• g2o https://github.com/RainerKuemmerle/g2o.git branch: master

In case of g2o, we advise you to build it from source. We would further advise you to use our modified robot_pose_ekf package

• navigation https://github.com/jhfrost/navigation.git branch: jade_devel

Accuracy of the occupancy grid map depends on

• the accuracy of the wheel odometry. We would advise you to add inertial measurements and use the robot_pose_ekf.
• the accuracy of your TF robot model. The transform base_link -> camera_link in particular must be accurate.

The Kinect has a small field of view. Consider using multiple cameras (e.g. front and backward facing cameras) for better performance. The algorithm supports an arbitrary number of cameras.