search

koide3 / glim

GLIM: versatile and extensible range-based 3D localization and mapping framework
MIT License
578 stars 77 forks source link

Loop closure doesn't fix the Z-axis drift. Any ideas? #108

Closed se7oluti0n closed 2 weeks ago

se7oluti0n commented 2 weeks ago

Hello,

First of all, thank you for the contribution. I have tested mapping on 1 round of the Park near my house. The park size is about 500m x 280m. I the the elevation of the Park is almost the same, some point on the trajectory has higher elevation, few meter.

For the map, there is one loop closure because I went 1 round of the park. I used offline viewer to create manual factor. But the final map after the loop-closure show huge different of elevation (about 30 meter) at two side of the park?

Any ideas to improve the z-drift after loop closure?

Screenshot from 2024-11-07 09-38-25 Screenshot from 2024-11-07 09-38-01

tahakenar commented 2 weeks ago

Hi, I am also facing the same issue with a relatively small map (~20m x ~170m road). I minimized cumulative Z-drift by manually closing the loop. Yet, there is still difference in Z axis between two ends of the map.

I provide the screenshots below. With no loop closure, Z-drift after closure and final map after loop closure respectively.

z-drift glim-drift-loop-closed glim-loop-closed

se7oluti0n commented 2 weeks ago

@tahakenar Thank for the information. One thing that is similar in our cases is we both traveled in a quite narrow road with the high tree at the sides at some points. May that case cause the error in the pitch angle estimation and lead to the z drift, I guess.

tahakenar commented 2 weeks ago

@se7oluti0n I processed my bag file to filter out point cloud data. I clamped points using boundaries -15m to 15m for x and y. In z axis I clamped between -1.5m to 2.5m. Seems like it yields better results in terms of z-axis drift.

Before finding overlapping submaps and manual loop closure:

glim-clamped-no-loop-closure

After manual loop closure and finding overlapping submaps:

glim-final

flat

skohlbr commented 2 weeks ago

Depends on the use case, but the https://github.com/koide3/glim_ext?tab=readme-ov-file#flat-earther plugin might solve this issue if the environment is approximately flat.

koide3 commented 2 weeks ago

Yes, the flat_earther plugin should solve the issue. In particular, if your environment is approximately flat overall, setting max_neighbor_distance in config_flat_earther.json to a large value (e.g., 1000 m) forces the map to be flat.

se7oluti0n commented 2 weeks ago

@skohlbr @koide3 Thanks for the suggestion. Closed topic here.