Mesh Generation Wrong Layout Result & Polishment

[Carton22]

Hi @RuanJY ,

fantastic work!
I manage to use your code to build mesh.
I am trying to drive my turtlebot 4 with Unitree L2 Lidar. I used your original code in a ROS1 docker.

I scanned around more then 2 rooms and 3 hallways, however the results does not looks good on my side.
The problem looks like the odometry of the Lidar shifted sometime.

Moreover, the generated mesh map layout is not the same as the layout in the real world.

Any suggestions or instructions on the possible reasons behind would be appreciated!

unilidar_2room_slamesh_result.mp4

[RuanJY]

when the valid contriaint is not enough in , especially with this sparse lidar, the pose may jump.

suggestion:
1, it seems that the map is not horizental, which is related to the instalation direction of sensor, you can change "correction_pitch_degree: -90" in the config file to adjust it
2, if the constraints is not enought, you can find the 'updated mesh' is sparse (shown as yellow triangle mesh in rviz), you can increase the "variance_register: 0.60"

[Carton22]

Thank you so much for your suggestions! @RuanJY
I did try to adjust these 2 parameters

For correction_pitch_degree, I changed to -90, 0 , 90. Does this parameter only affect how the viewing angle in rviz, or it would also affect how the mesh generates?

For variance_register, I changed to 0.8, 1, 5, the pose would still jump at some point.

I wonder have you tested using Unitree Lidar and run the robot in the physical world, do you encounter this kind of jump issue?
Or all your demos are tested using Ouster Lidar?

PS: I tested all in indoor environment in my office building.

Thanks again for any help~

[Carton22]

I used the same hardware setup and run the point_lio_ros2, it managed to scan the whole floor of my building. And I did not find too obvious jump or shift of the IMU/transform of the unitree L2 lidar.

Below is the result of the point_lio_ros2, it is a point cloud .ply model. I think point_lio_ros2 use IMU of Unitree Lidar as one of the inputs

hope this can help find out the issue.

[Carton22]

For your reference, this is how SLAMesh result looks like with Unitree LiDAR L2 indoor ros bag playing.

I am not pretty sure if the result is correct or not. Looks like it is correct, but not sure if the mesh results is overlapped and repeatedly generated in some areas.