Algorithm utilizes the MRPT mrpt-graphslam library to execute single, multi-robot graphSLAM using information from odometry and LaserScan ROS topics.
mrpt_graphslam_2d is heavily based on the native MRPT graphslam-engine application. Furthermore the command-line arguments offered by the latter can be provided by the user as parameters in the ROS Parameter Server. See the provided roslaunch files for examples of this. Also see the API of mrpt_graphslam_2d available here
Single robot algorithm
Real-Time experiment - short loop
A sample rosbag is included in the
rosbags/demo_short_loopdirectory. To run this just launch the sr_graphslam_demo.launch file:
roslaunch mrpt_graphslam_2d sr_graphslam_demo.launch start_rviz:=True
Demo rosbag contains Laser Scan measurements (and odometry which is not usuable in the algorithm due to the topic type)
Real-Time experiment - ground-truth data included
A real-time experiment using a KUKA youbot and having a rough ground-truth estimate path is available here. Ground-Truth estimation is provided using 2 ps3 cameras placed on the room ceiling. The cameras using the ar_sys ROS package are tracking Aruco markers that act as the workspace origin (static marker) and as the robot executing SLAM (moving marker) respectively. To run this demo, simply download the entire folder, place it directly under the
rosbagsdirectory of the mrpt_graphslam_2d package and run the sr_graphslam_demo_gt.launch file.
roslaunch mrpt_graphslam_2d sr_graphslam_demo_gt.launch
One can also tinker with the aforementioned launchfile to enable/disable the different visualization features, use another decider/optimizer class, or modify the .ini file to change the algorithm behavior.
Robot movement starts after ~60''. Due to different timestamps in the laserscans, odometry topics the algorithm feedback is lagging a bit compared to the ground-truth visualization
sr_graphslam_demo_gt file uses the pass_all_args XML directive which is only available starting from the ROS Jade distribution. If you have an older one, just replace this with all the args in that file exclusively. For an example on how to do this, see the sr_graphslam_demo.launch file
Rosbag was recorded in the Mechanical Engineering School of the National Technical University of Athens
Multi-robot support is provided via the csl_mr_slam package suite as well as the use of multimaster_fkie packages utilized for communication across the various ROS agents.
Multi-robot real-time experiments - rosbag
A rosbag collected from real-robots can be found in the following link (more rosbags to be added in the MRPT-2.0 branch):
To run a demo using one of the corresponding rosbags download that directory,
place it directly under the
rosbags directory of the mrpt_graphslam_2d
package and run the
run_mr_graphslam_demo.sh script of the
csl_robots_gazebo package. The latter script (as is standard with the nodes of the
csl_robots_gazebo package reads its configuration parameters off the shell
environment at hand and adjusts its behavior accordingly. As an examaple users
can use the following command for running a multi-robot demo.
# see the script for configuration variables rosrun csl_robots_gazebo run_mr_graphslam_demo.sh
For this to work, you need to have the csl_mr_slam package suite (as well as the packages that the latter depends on) in your catkin workspace.
When executing multi-robot graphSLAM using either Gazebo (via the
csl_robots_gazebopackage or using measurements from the prerecorded rosbags, you have to, a priori, generate the necessary rviz files from the template files found in
$(rospack find mrpt_graphslam_2d)/rviz/templates. To do that you have to run the
$(rospack find mrpt_graphslam_2d)/nodes/rename_rviz_topics.pyscript which changes the necessary topic names based on the running computer's hostname. See documentation of the latter script for more on its usage.
Multi-robot simulations are supported in the Gazebo Simulator via the csl_hw_setup ROS package. An example of running such a simulation is given below.
A complete example of executing multi-robot graphSLAM in the Gazebo simulation environment is presented in the following video:
Real-time multi-robot experiments
As with its single-robot variant, support for running the algorithm in a real-time multi-robot setup is provided via the csl_mr_slam/csl_hw_setup package.
An example of executing real-time multi-robot graphSLAM is given here.