FSD Localization package

This package implements the Monte Carlo Localization technique using Free-Space Density (FSD) as observation model, as proposed by Renan Maffei et al. [1]

1. Installation

1. This package works in ROS melodic (and kinetic). For instructions on how to install and configure a ROS workspace see http://wiki.ros.org/ROS/Tutorials/InstallingandConfiguringROSEnvironment

   • The framework has two dependencies that must be installed

     • Glut: graphical user interface
     • FreeImage: screenshot saving

     sudo apt-get install freeglut3-dev libfreeimage-dev

2. Copy the folder of 'fsd_localization' package into the src folder. It should look like this:

    ROS_WORKSPACE_PATH/
        build/
        devel/
        src/
          fsd_localization/

3. To compile the package, just type catkin_make inside the main folder of the workspace.
   catkin_make

2. Running the program

The program can be executed as follows
rosrun fsd_localization fsd_localization

Parameters:

• mapName: The program loads a 2d grid map (in .txt format). The map should be located in the 'maps' folder within the package.
  This work was developed using data from the Robot@home dataset, converted to the ROS format using the robotathome_at_ros package, available at https://github.com/phir2-lab/robotathome_at_ros.
  The default option is alma. Other options are anto, pare, rx2. All these maps are contained in the 'maps' folder.

• method: Different particle weighting strategies are implemented in the MCL.cpp file, and their use can be selected with this parameter.

  • ABS_PCL_FSD: Absolute FSD - Free-Space Density using depth information obtained by RGB-D cameras;
  • INT_PCL_FSD (default): Interval FSD - An Interval-based extension of the previous approach that computes the Free-Space Density using not only the known free-space, but also the possible free-space in unknown regions;
  • CLikelihood: scan matching between K random measurements from the point cloud, analyzing beam endpoints using a likelihood map, as used in [2]
  • CRaycast: similar to the previous method, but performing raycasting to obtain more precise results;
  • LRaycast: scan matching using the 2D laser readings;
  • Motion: No observation model - only removing particles that go over obstacles or outside the map

• datasetType: There are two option: multi_cam or single_cam (default). The first option uses the 4 RGB-D cameras of the Robot@Home dataset, and the second uses only the frontal RGB-D camera.

Ex:
rosrun fsd_localization fsd_localization _mapName:="anto" _method:="ABS_PCL_FSD"

3. Running everything together with a launch file

In the 'launch' folder is available a launch file to simultaneously run the program, the RViz viewer and bagfiles with data from the Robot@home dataset converted to the ROS format using the robotathome_at_ros package.

For proper operation, you need to install beforehand the robotathome_at_ros package in the same ROS workspace and generate the bagfiles following the package instructions.

To run the launch file, type
roslaunch fsd_localization run_MCL.launch

The same three parameters of the FSD localization program (mapName, method, datasetType) can be set by the user in the launch file.

Ex:
roslaunch fsd_localization run_MCL.launch mapName:="rx2" datasetType:="single_cam"

4. Framework controls

w, a, s, d : scrolls viewed area

+, -: changes zoom

v, b: changes view mode

ESC: close program

References

[1] Maffei, R. Q., Pittol, D., Mantelli, M., Prestes, E. Kolberg, M. Global Localization Over 2D Floor Plans with Free-Space Density Based on Depth Information. Accepted in: 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Las Vegas, USA. 2020. (Accepted version)

[2] W. Winterhalter, F. Fleckenstein, B. Steder, L. Spinello, and W. Burgard, “Accurate indoor localization for rgb-d smartphones and tablets given 2d floor plans,” in Proc. of IROS, 2015, pp. 3138–3143.