SAGE-ICP

Paper   •   Arxiv   •   YouTube   •   bilibili

SAGE-ICP is a semantic information-assisted point-to-point ICP (ICRA 2024, accepted).

Jiaming Cui, Jiming Chen, Liang Li

---

Prerequisites

1. ROS2
2. Dependencies

sudo apt-get install ros-<ros_distro>-tf-transformations
pip install transforms3d kiss-icp lark-parser

3. Semantic segmentation models for LiDAR point clouds, could select one installation.
   Cylinder3D, RangeNet++

Install

mkdir -p ~/sage-icp/src && cd ~/sage-icp/src
git clone https://github.com/NeSC-IV/sage-icp.git
cd .. && colcon build --symlink-install

Evaluation

To evaluate cylinder3D on KITTI Odometry, please run:

source install/setup.bash
ros2 launch sage_icp sem_odom.launch.py

To evaluate RangeNet++ on KITTI Odometry, please run:

source install/setup.bash
ros2 launch sage_icp range_odom.launch.py

Citation

If you use this library for any academic work, please cite our original Paper.

@INPROCEEDINGS{10610280,
  author={Cui, Jiaming and Chen, Jiming and Li, Liang},
  booktitle={2024 IEEE International Conference on Robotics and Automation (ICRA)}, 
  title={SAGE-ICP: Semantic Information-Assisted ICP}, 
  year={2024},
  volume={},
  number={},
  pages={8537-8543},
  keywords={Location awareness;Training;Laser radar;Accuracy;Semantic segmentation;Semantics;Pose estimation},
  doi={10.1109/ICRA57147.2024.10610280}}

Acknowledgements

This repo is built upon KISS-ICP, Cylinder3D and RangeNet++.
This project is free software made available under the MIT License. For details see the LICENSE file.