A highly efficient, CUDA-accelerated framework for Fast, Accurate, and Robust LiDAR-Inertial Odometry.

📖 Documentation

Installation

Pull the pre-built CUDA image from the GitHub Container Registry:

docker pull ghcr.io/tumftm/far-lio:latest

Or build it yourself:

git clone --recursive https://github.com/TUMFTM/FAR-LIO.git
cd FAR-LIO
docker build -f docker/Dockerfile -t ghcr.io/tumftm/far-lio:latest .

See the installation guide for prerequisites and options.

Usage

Run FAR-LIO on its own (e.g. with live sensors — needs a GPU):

docker compose --profile far-lio up

Or replay a ROS 2 bag through it — the helper script mounts the bag and starts both services:

./run.sh /path/to/rosbag # bag directory (metadata.yaml) or an .mcap file

ROS_DOMAIN_ID is taken from your environment if set, otherwise 0. See the usage guide for details.

Racetrack Deployment (A2RL)

An excerpt of FAR-LIO running on an autonomous race car on a racetrack as part of the Abu Dhabi Autonomous Racing League (A2RL).

_{Onboard camera footage}

_{FAR-LIO's odometry}

See the documentation for the architecture and further application domains.

References

If you use FAR-LIO in your research, please cite our paper:

@misc{2026far-lio,
      title={FAR-LIO: Enabling High-Speed Autonomy through Fast, Accurate, and Robust LiDAR-Inertial Odometry}, 
      author={Maximilian Leitenstern and Marcel Weinmann and Patrick Haft and Tobias Lasser and Dominik Kulmer and Markus Lienkamp},
      year={2026},
      eprint={2606.26010},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2606.26010},
}

Core Developers

Marcel Weinmann
Maximilian Leitenstern
Institute of Automotive Technology, School of Engineering and Design, Technical University of Munich, 85748 Garching, Germany

Acknowledgements

We thank Patrick Haft and Tobias Lasser (NVIDIA Corporation) for their assistance during the CUDA development.