GroundSeg3D

[haider8645]

GroundSeg3D is a sensor and vehicle agnostic ground segmentation algorithm designed for all-terrain use. This method demonstrates robust performance across various open-source datasets, including SemanticKITTI, RELLIS-3D, UrbanNav, DOALS, and BotanicGarden, without needing parameter tuning. A video preview highlights its adaptability and effectiveness. The corresponding research paper has been submitted for ICRA 2025, and the code will be made open-source post-publication.

Required Msg Inputs:

1. Pointcloud2
2. IMU

Runtimes:

Could this be of interest to autoware community?

[amadeuszsz]

@haider8645

Thanks for reaching our community. The demo itself looks great! We are open for contribution from your side. Referring directly to your project, I think it will match to our perception pipeline and in case of better performance, it can replace current solution for ground segmentation.

[haider8645]

Thanks for the response. Alright then I will look into the steps needed for an integration of the software in the perception pipeline