We propose LCR-Net to tackle both LiDAR-based loop closing and relocalization. It exploits novel feature extraction and a pose-aware attention mechanism to precisely estimate similarities and 6-DoF poses between pairs of LiDAR scans.
LCR-Net has been integrated into SLAM systems and achieves robust and accurate online LiDAR SLAM in outdoor driving environments.
The code is coming soon!
This work has been accepted by IEEE Transactions on Robotics (TRO) 🎉 [pdf] [video]. Please cite our work using:
@article{shi2024lcrnet,
author={Shi, Chenghao and Chen, Xieyuanli and Xiao, Junhao and Dai, Bin and Lu, Huimin},
journal={IEEE Transactions on Robotics (TRO)},
title={Fast and Accurate Deep Loop Closing and Relocalization for Reliable LiDAR SLAM},
year={2024},
}