map drift a lot,loop not found

[liuyf5231]

Hi koide3 ,

I tested hdl graph slam by robosense 16 lidar, but i found that there is something wrong with the map.

--- loop detection ---
num_candidates: 8
matching........ done
best_score: 36.815    time: 0.000[sec]
loop not found...

tf between map and odom,

At time 1572313024.112
- Translation: [-5.383, -2.814, 16.107]
- Rotation: in Quaternion [-0.001, -0.031, -0.034, 0.999]
            in RPY (radian) [-0.001, -0.062, -0.068]
            in RPY (degree) [-0.034, -3.564, -3.881]

[liuyf5231]

[liuyf5231]

[koide3]

Hi @liuyf5231 , NDT-based methods may not work well with 16-line LIDARs because it is not very robust to sparse point clouds. Could you try to use GICP or GICP_OMP instead of NDT_OMP (by changing parameters in the launch file)? If you are using ROS melodic, GICP should work well (PCL on Kinetic or earlier has a bug). You can also try to slow down the rosbag playback speed: bash rosbag play --clock -r 0.2 bagfilename.bag

[liuyf5231]

Hi @liuyf5231 , NDT-based methods may not work well with 16-line LIDARs because it is not very robust to sparse point clouds. Could you try to use GICP or GICP_OMP instead of NDT_OMP (by changing parameters in the launch file)? If you are using ROS melodic, GICP should work well (PCL on Kinetic or earlier has a bug). You can also try to slow down the rosbag playback speed: bash rosbag play --clock -r 0.2 bagfilename.bag

It works! if I want to use hdl localization,I need to change the registration method to GICP_OMP?

[koide3]

In that case, maybe you can use NDT_OMP since the target point cloud (map cloud) is dense.