3D Mapping of a Collapsed Car Parking Lot

The following data was collected during NIST Response Robot Evaluation Exercise 2008 at Disaster City, Texas. The related map was generated using 3D Plane SLAM.

Scenario Data

Photos from the Site

Front view of the crash site. Close up. The robot collecting data.

Raw Data

3D Plane SLAM Results

Below some videos are shown with animated results of the 3D mapping with 3D Plane SLAM. Please click the images to play the movies.


Front Camera View 3D model before relaxation
Front camera view during data-collection (17.5 MB) Model after only plane-matching (23 MB) range images as movie (0.5 MB)
Polygonalized Soeren Registered point-clouds before Registered point-clouds after
A plane-fitted and polygonized human (11 MB) Map as point-clouds using the registration result of plane-matching only (483 MB) Improved map as point-clouds using the registration result of plane-matching and relaxation (360 MB)


Plane-SLAM Maps in X3D format

The results of the Plane-SLAM are also available as X3D data. Use "tar zxvf file-name" to uncompress and untar into folders, and then load main.x3d in the output folder using an X3D viewer.

  • After pairwise plane-matching, but before relaxation:
    • Map as planar-patches (0.6 MB) includes the loop-closing edges; This is easier to visualize compared to the point-cloud maps.
    • Map as point-clouds (15.5 MB) ; In both cases the registration was done using plane-matching. Point-clouds subsampled by 3 for easier visualization.
  • After pairwise plane-matching followed by loop-closing and relaxation:


Comparison to ICP

3D-Plane-SLAM outperforms Iterative Closest Point (ICP) in this scenario in terms of computation speed and robustness. Due to the lack of meaningful odometry, ICP fails with several pairwise registrations.

video of the broken 3D map generated by ICP (full version: 413 MB) (downscaled version: 20 MB)


Freely available X3D viewer