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.
The dwelling is a set taken in a house were a major flooding has taken place. There are quite some amounts of rubble in the scenario. Parts of the house are destroyed. Especially, the ceiling is collapsed in some rooms. An accelerated movie from the robot's front camera gives an impression of the scenario. Below, a few pictures from the dwelling are also shown.
The raw data of the dwelling can be downloaded via the following links
- 3D scans:
- as raw point clouds: http://robotics.jacobs-university.de/datasets/RAW/RREE08-dwelling-pcl.tgz (41 MB)
- as png range images: http://robotics.jacobs-university.de/datasets/RAW/RREE08-dwelling-range-png.tgz (848 KB)
- movie of the range images: http://robotics.jacobs-university.de/multimedia/movies/RREE08-Range-Dwelling.mpg (948 KB)
- front camera images:
- all raw images as jpegs: http://robotics.jacobs-university.de/datasets/RAW/RREE08-dwelling-frontcamera.zip (120 MB)
- the images at the locations where the scans were taken: http://robotics.jacobs-university.de/datasets/RAW/RREE08-dwelling-frontcamera-selected.zip (862 KB)
- front camera as video: http://robotics.jacobs-university.de/multimedia/movies/RREE08-dwelling-frontcam.avi (28 MB)
The following 3D map was generated using 3D Plane SLAM and 70 scans from this scenario.
views inside the building:
(click images to enlarge them)
Performance Comparison to ICP
Experiments with Iterative Closest Point (ICP) for 3D point cloud registration did not lead to meaningful results in this scenario. One explanation is the lack of motion estimates due to the lack of odometry, respectively the errors in the odometry – using the recorded odometry data in this scenario makes the ICP results even worse. There are hence no reasonable initial guesses for ICP as starting conditions. Furthermore, the robot moves through several small doors. Larger parts of a room visible in one scan hence disappear in the scan where mainly the new room is covered. This also seemed to pose problems for ICP. Below an animation of the "map" generated by ICP:
Animation of the "map" generated by ICP (click image to play the video; 214 KB)