Application: 2d-slam-demo

1. Description

This GUI application is a demonstration of 2D range-bearing SLAM under the paradigm of EKF-based SLAM, in a didactic way that allows extensive experimentation with data-association and parameter tuning.

It is an extension to a similar Matlab program developed by J. Neira and J. D. Tardós (University of Zaragoza).

At the core of this program lies one MRPT class: mrpt::slam::CRangeBearingKFSLAM2D.

The paper explaining the Matching Likelihood criteron for data association (one of the two choices offered in this program) is [BGonzalezJimenezFernandezM12] (BibTex, Draft PDF ).

2. Demo video

3. Command line options

Invoking the program without any flag will launch it in normal, interactive, mode:

USAGE:

2d-slam-demo  [-r] [-n] [-c <params.ini>] [--] [--version] [-h]


Where:

-r,  --norun
    Just load the config file, don't run it.

-n,  --nogui
    Don't stay in the GUI, exit after the experiment.

-c <params.ini>,  --config <params.ini>
    Config file to load

--,  --ignore_rest
    Ignores the rest of the labeled arguments following this flag.

--version
    Displays version information and exits.

-h,  --help
    Displays usage information and exits.


2d-slam-demo

With --config, a non-interactive experiment is run first (its statistics are printed and appended to slam_out.txt); with --nogui the program then exits, without ever opening a window, so batch experiments can run on machines without a display.

4. User interface

All views are dockable panels; their layout is remembered across sessions.

  • Ground truth: the landmarks, the robot and the area seen by its sensor.

  • Observation: the last observation in the robot frame, with the 3-sigma ellipses of each reading.

  • Estimated map: the robot and landmarks estimated by the EKF, with their 3-sigma ellipses (optionally labeled with their real landmark index).

  • Individual compatibility: observations and predictions in range-bearing space, with the individually compatible pairs and the final associations; and the DA matrix of their distances.

  • Errors: robot pose errors in x, y and heading vs. step, with their 3-sigma bounds.

  • Statistics: data association results (true/false positives and negatives, JCBB iterations) and computation time per step.

  • Parameters: every parameter of the KF, the robot motion, the sensor, the data association and the ground truth map (square corridor, random, or a text file with one “x y” landmark per row), editable while the simulation runs.

Toolbar and keyboard: Reset (R), Step (S), Run/Stop (Space) and Batch (B, a full non-interactive run in the background, which can be cancelled). The Tools menu loads and saves the parameters, saves the filter state or the last data association, records the simulated data to a rawlog file, and shows the profiler statistics.