search

hyye / lio-mapping

Implementation of Tightly Coupled 3D Lidar Inertial Odometry and Mapping (LIO-mapping)
https://sites.google.com/view/lio-mapping
GNU General Public License v3.0
898 stars 320 forks source link

Compatibility with Google Cartographer datasets #19

Closed andre-nguyen closed 4 years ago

andre-nguyen commented 4 years ago

Hi @hyye ,

How would you adapt lio-mapping to work with the google cartographer datasets? I see they converted each lidar packets to a pointcloud while you use the accumulated clouds from the velodyne driver. Other than reassembling clouds, are there any parameters that neet to be adapted?

Thanks,

hyye commented 4 years ago

Hi @andre-nguyen, I haven't tried to reassemble the clouds from the cartographer datasets yet.

I think in reassembling, we need to make the cloud as a full sweep and set the timestamp of the cloud as the middle point's. The Velodyne's ROS driver may be a useful reference.

andre-nguyen commented 4 years ago

Thanks, I stamped with the end of the cloud and it seemed to work. Will look into which time stamp works best. Did you use the fully assembled clouds from the ros driver?

andre-nguyen commented 4 years ago

https://github.com/ros-drivers/velodyne/blob/master/velodyne_driver/src/driver/driver.cc#L252

Looks like they stamp with the last timestamp. I find it interesting that lio-mapping works so well when the timing isn't particularly tight.

hyye commented 4 years ago

Glad to hear it works! I used the fully assembled one from the ros driver. I got it wrong; It should be the end of the cloud. I transformed the points to the middle time in the code.

andre-nguyen commented 4 years ago

Unfortunately, it crashes on some of the datasets due to some kind of kdtree error. It is unclear to me why. However, I have my own dataset that is formatted like the cartographer ones and lio seems to work on mine.