OverlapNet

OverlapNet

Loop closing for 3D LiDAR SLAM

Description

When a robot returns to its start, drift makes the SLAM map misalign; the key is recognizing a place it has seen. OverlapNet from the University of Bonn is a deep network for loop closing in 3D LiDAR SLAM that estimates overlap and relative yaw between scans.

It improves LiDAR SLAM loop closure, published at RSS 2020.

Features



Loop closure:Revisited places.

Overlap:Scan similarity.

Yaw:Relative heading.