koide3 / hdl_graph_slam

3D LIDAR-based Graph SLAM
BSD 2-Clause "Simplified" License
2.02k stars 730 forks source link

Scan matching optimization doesn't seem very effective #241

Open MatCapLynx opened 2 years ago

MatCapLynx commented 2 years ago

Hi, thanks for the amazing work. I've been using your package and it has proven very useful.

I need to perform SLAM in a challenging environnment where scan matching can not produce a relevant odometry. So instead of using the scan matching odometry nodelet, I'm using my own odometry. This system is working quite well but after a while it starts to drift. This drift could be corrected by scan matching during pose graph optimization. I've tried to tweak scan matching and odometry edge parameters but I haven't reached any visible result. Could you precise which parameters are the most important for this matter ? maybe I'm missing some.

I know this can work because I've successfully optimized my map using the interactive_slam package and its Edge refinement tool. I've tried to use the same parameters but it doesn't change anything. While using hdl_graph_slam I've checked the /hdl_graph_slam/odom2pub tf topic which is the difference between odometry and optimized position. There was almost no difference (1E-8m) as if optimization wasn't doing anything.

Thanks for the help.