Comparing Open Source SLAM Algorithms#
In order to compare SLAM algorithms their trajectory errors can be compared in terms of positions and orientations, thus ground truth should be generated and GNSS/INS can be used for this purpose.
Setup#
We have used a mapping kit which contains Velodyne VLP16 LiDAR and Applanix LVX GNSS/INS sensors. With using this kit we could collect data for ground truth and SLAM algorithms simultaneously. Since ground truth is generated by GNSS/INS, it should be transformed to LiDAR frame. Additionally, average positional error in the postprocessed GNSS/INS data is 0.06 meters.
Transform#
To calculate error's of slam algorithms ground truth trajectory and slam trajectories must be in the same frame. Hence, whether ground truth or slam trajectory should rotate and translate to the other's frame.
Tf tree of this transform is:
To perform necessary rotation and translation firstly transform matrices are calculated then transforms applied on GNSS/INS positions by using calculated transform matrices.
Before transformation:
Results#
RViz output of both of the ground truth and slam trajectories and the graphs of the positional errors in meters of the slam algorithms are depicted below.
| SLAM Algorithm | Trajectory After Transformation | Position Error Graphs | Average Position Error (meters) |
|---|---|---|---|
| hdl_graph_slam |  |
 |
x:11.903188 y: 3.887582 z: 13.905123 |
| lidarslam_ros2 |  |
 |
x: 8.949974 y: 2.818557 z: 16.495759 |
TODO#
- Transform between GNSS/INS trajectory and SLAM trajectory will be implemented using ros tf.
- SLAM algorithms tested with IMU data.
- Remaining SLAM algorithms will be tested.