Request for Assistance with Parameter Tuning in Patchwork++ for Ground Segmentation in ROS1

[wanghuohuo0716]

Hi, Your work on Patchwork++ is outstanding, and I want to thank you very much for open-sourcing this project.

I tried using Patchwork++ in ROS1 to process data from two scenarios, but I found that it couldn't accurately segment the ground in slope scenario.

Can Patchwork++ handle these types of scenarios, and would it require parameter adjustments? I hope to get your assistance on this.

Could you please share some tips or experiences on tuning the parameters?

Scenario 1: An 80-line LiDAR (without intensity) with people, uphill slopes, and curbs (about 15 cm high), the robot moves forward in the negative Y-axis direction of the LiDAR.

Expected behavior: Points other than the curbs (those whose height is generally not lower than the curb) and all points of people should be identified as non-ground, while the uphill slopes should be identified as ground points. You can clearly see two straight lines (the curbs) along the sides of the road.

Below are my dataset and parameter file.
pointcloud topic is '/hesai/pandar_points', Fixed Frame is 'PandarSwift'
scenario_1.zip

This is a picture of the real scenario:

The test results are as follows:

Scenario 2: A 16-line LiDAR (with intensity) on a road with uphill and downhill slopes, and curbs (about 20 cm high), the robot moves forward in the positive X-axis direction of the LiDAR..

Expected behavior: Points other than the curbs (those whose height is generally not lower than the curb) should be identified as non-ground, and the uphill and downhill slopes should be identified as ground points. You should clearly see two straight lines (the curbs) along the sides of the road as you walk along.

Below are my dataset and parameter file.
pointcloud topic is '/velodyne_points', Fixed Frame is 'bigrobot2/vehicle_frame'
scenario_2.zip

The test results are as follows: