Ground Blank Area and Human-Induced Obstacles in Large-Scale Outdoor Mapping with Unitree Go2 Robot Dog and Hesai XT16 LiDAR

[beihaiyisuren-cmyk]

I am currently using the Unitree EDU Go2 robot dog, with a Hesai XT16 LiDAR horizontally mounted on top of the robot dog. The overall image has been deployed. However, when I perform large-scale outdoor mapping, I noticed that the ground directly beneath the robot dog is blank. For example, when I walk in a circle, there is a circular blank area on the ground, while the surrounding areas are fine. Additionally, since I need to walk alongside the robot dog with my laptop, I find that people (including myself) introduce many obstacles into the map. I would appreciate your response. Thank you.

[beihaiyisuren-cmyk]

[tsengapola]

Have you enabled patch_first_ring_to_baselink to true?

dddmr_navigation/src/dddmr_lego_loam/lego_loam_bor/config/loam_bag_c16_config.yaml

Line 34 in 67c9a46

patch_first_ring_to_baselink: true

If it is already true, launch the slam and do not move the robot, start tuning your ground FOV parameter:

dddmr_navigation/src/dddmr_lego_loam/lego_loam_bor/config/loam_bag_c16_config.yaml

Line 36 in 67c9a46

ground_fov_bottom: -0.2617994 # -15deg

Make sure your ground of your first frame is not hollow, and then start to move the robot slowly to see is the ground being constructed perfectly.

Regarding people in the map, since your lidar is only 16 lines, the deep learning approach will not work. The effective way is to use point cloud editor to remove those points.
We will quickly implement point cloud editor as early as possible.

[tsengapola]

@beihaiyisuren-cmyk we release point cloud deleting tools!
https://github.com/dfl-rlab/dddmr_navigation/tree/main/src/dddmr_perception_3d#delete-point-cloud-in-pcd-file