Parameter tuning suggestions

vectr-ucla / direct_lidar_odometry

[IEEE RA-L & ICRA'22] A lightweight and computationally-efficient frontend LiDAR odometry solution with consistent and accurate localization.

MIT License

882 stars 186 forks source link

Parameter tuning suggestions #35

Closed JereKnuutinen closed 1 year ago

JereKnuutinen commented 1 year ago

I am doing research related to motion based calibration and I am using DLO for lidar odometry generation. It has worked fine so far but with my current dataset it gives non reasonable pose estimates, what are the first parameters try to tune? Figure below shows some things related to my dataset. So it is infinity shape path in a planar ground. googlemaps_fig

sample pointclouds are like figure below. I have combined two vlp16 point clouds: sample_pointcloud

kennyjchen commented 1 year ago

Hi @JereKnuutinen -- thanks for using our work. I'll need more information about your dataset and how the pose estimates are not reasonable (or a sample bag). Is it jittery, or does localization not work altogether? Are you using LiDAR only or do you also have an IMU?

As for some parameters you can try tuning, check out this issue. Let me know if that helps!

JereKnuutinen commented 1 year ago

Hi, thanks for your reply! I am not using IMU only combined pointcloud from two vlp 16 lidar sensors. Here is a rosbag if you want to take a look. Also parameters from previous issue do not worked.

https://drive.google.com/file/d/1I2svjYyjBIBrcEF_-biIWNRh4FgDZllM/view?usp=sharing

Topic name is \merged_pc

kennyjchen commented 1 year ago

Have you tried it using just one point cloud (without merging)? How does that look?

kennyjchen commented 1 year ago

Ah -- I see the problem after playing with the data you sent. Two issues:

There are multiple points on the body of the robot (car?), which can cause issues in scan-matching

You're using a very sparse, long-range LiDAR. So the current scan-to-map GICP threshold is too small for that (converges to a local minima)

Try the following:

Set cropBoxFilter: size to 10 (or some value to get rid points on your platform) in params.yaml
Set gicp: s2m: maxCorrespondenceDistance to 10 (play around with this) in params.yaml

Let me know if that works.

JereKnuutinen commented 1 year ago

Thank you very much for your help, it seems to be working now!

kennyjchen commented 1 year ago

Nice! Glad to hear that.