Re: Tilted 3D map with floam

Posted by matlabbe on
URL: http://official-rtab-map-forum.206.s1.nabble.com/Tilted-3D-map-with-floam-tp10380p10463.html


Hi,

In my test, the odometry will show only about 5 cm error in z between the beginning and the end of the trajectory. Also, I ignored /tf and /tf_static from the bag when replaying, and just estimate directly in os_sensor. I didn't read the whole issue, but it seems that lidar scans are projected in os_sensor instead os_lidar frame. That should not be a problem if the resulting xyz point cloud is in os_sensor frame. That could be a problem if the lidar scans are projected in os_lidar frame, but published point cloud has os_sensor frame_id instead. For the IMU frame, an experiment could be to stick another IMU on the lidar, and check if the values are aligned together.

If you tried exactly what I did with the rosbag and results are that different, maybe the rtabmap version is different. If you are using the ros binaries, that should be similar. If you built from source, make sure rtabmap library is built with libpointmatcher. Another thing that could be different is the computation power that may be lower than on computer I tested. If so, you can increase "Icp/VoxelSize" and "OdomF2M/ScanSubtractRadius" parameters. I think you can also remove "Icp/Force4DoF: true", probably forgot to remove it after testing with imu.

If you just started the mapping, icp_odometry should return values around (0,0,0) or exactly (0,0,0) for the first frame. An error in URDF would not show up there. Doublecheck that you don't have topics published from two different sources on same topic name. For example, I don't remember if the rosbag you sent had already some odometry topics in it, but if you are testing with a bag, make sure to remap the topics in the bag to some random name to avoid confusion with topics regenerated.

cheers,
Mathieu