RKO-LIO

Robust LiDAR-Inertial Odometry Without Sensor-Specific Modelling

Four different platforms, four different environments, one odometry system

This repository contains a ROS1 (catkin / roscpp) port of RKO-LIO, intended for Ubuntu 22.04 with the ros-one distribution.

Quick Start (ROS1)

Prerequisites

• Ubuntu 22.04
• ros-one (ROS 1) with roscpp, rosbag, tf2, sensor_msgs, nav_msgs
• A catkin workspace (e.g. catkin_tools)

System packages commonly required for building:

sudo apt install libeigen3-dev libtbb-dev

Build

Clone this package into your catkin workspace src/ directory, then:

cd /path/to/your/catkin_ws
rosdep install --from-paths src --ignore-src -r -y
catkin build rko_lio --cmake-args -DCMAKE_BUILD_TYPE=Release
source devel/setup.bash

Build in Release mode (-DCMAKE_BUILD_TYPE=Release). Debug builds are significantly slower and may not keep up with 10 Hz LiDAR in real time.

If rosdep cannot resolve dependencies such as Sophus or tsl-robin-map, fetch them automatically at configure time:

catkin build rko_lio --cmake-args -DCMAKE_BUILD_TYPE=Release -DRKO_LIO_FETCH_CONTENT_DEPS=ON

Online odometry

The launch file loads defaults from config/ros_default.yaml. Override sensor topics and frames as needed:

roslaunch rko_lio odometry.launch \
  lidar_topic:=/sensing/pandar \
  base_frame:=base_link \
  rviz:=false

imu_topic, lidar_topic, and base_frame must be set either via launch arguments or in the YAML config loaded by config_file. When using ros_default.yaml, imu_topic defaults to /sensing/ins/imu.

Example with an explicit config file:

roslaunch rko_lio odometry.launch \
  config_file:=$(rospack find rko_lio)/config/ros_default.yaml \
  lidar_topic:=/sensing/pandar \
  base_frame:=base_link

Launch modes

odometry.launch supports three modes via the mode argument:

mode	Node	Description
online (default)	online_node	Threaded async registration from live topics
online_imu_rate	online_imu_rate_node	Sequential pipeline; optional IMU-rate odometry
offline	offline_node	Replay from a rosbag (bag_path required)

Offline example:

roslaunch rko_lio odometry.launch \
  mode:=offline \
  bag_path:=/path/to/data.bag \
  lidar_topic:=/sensing/pandar \
  base_frame:=base_link \
  rviz:=false

Published topics (defaults)

Topic	Type	Description
/rko_lio/odom	nav_msgs/Odometry	LiDAR-rate odometry
rko_lio/lidar_acceleration	custom	LiDAR-frame acceleration estimate
/rko_lio/local_map	sensor_msgs/PointCloud2	Local map (if publish_local_map:=true)

TF: odom → base_link (child frame configurable via base_frame).

Configuration

Parameters are loaded onto the private namespace /rko_lio/ from YAML. See config/ros_default.yaml for defaults and available LIO tuning options (voxel_size, max_range, deskew, etc.).

To inspect all launch arguments:

roslaunch rko_lio odometry.launch --help

Extrinsics and convention

The system needs extrinsics between IMU, LiDAR, and the base frame. If your TF tree defines imu → base_link and pandar → base_link (or equivalent), extrinsics are resolved automatically from TF.

Otherwise, set them in the config YAML:

extrinsic_imu2base_quat_xyzw_xyz: [qx, qy, qz, qw, tx, ty, tz]
extrinsic_lidar2base_quat_xyzw_xyz: [qx, qy, qz, qw, tx, ty, tz]

Each value is a list of seven numbers: quaternion (x, y, z, w) followed by translation (x, y, z). Example identity: [0, 0, 0, 1, 0, 0, 0].

Transform naming follows extrinsic_<from>2<to>: a vector expressed in <from> is transformed into <to>.

Notes on real-time operation

If registration is slightly slower than the LiDAR rate, you may see throttled warnings such as Registration backlog: skipping N stale lidar scan(s). This is expected under load; the node keeps the latest scan and continues publishing odometry. For best throughput, always build with -DCMAKE_BUILD_TYPE=Release.

Python (optional)

The upstream project also provides a standalone Python CLI and PyPI package. That path is independent of the ROS1 nodes above.

pip install "rko_lio[all]"
rko_lio -v /path/to/data

See rko_lio --help and the upstream Python docs.

Citation

If you found this work useful, please consider leaving a star ⭐ on this repository and citing our paper (RA-L | arXiv):

@article{malladi2026ral,
  author      = {M.V.R. Malladi and T. Guadagnino and L. Lobefaro and C. Stachniss},
  title       = {A Robust Approach for LiDAR-Inertial Odometry Without Sensor-Specific Modeling},
  journal     = {IEEE Robotics and Automation Letters},
  year        = {2026},
  volume      = {11},
  number      = {6},
  pages       = {7420--7427},
  doi         = {10.1109/LRA.2026.3685966},
}

The paper supplementary material page collects extra figures and explanations pertaining to the same paper.

You can check out the branch ral_submission for the version of the code used for submission to RA-L. Please note that that branch is meant to be an as-is reproduction of the code used during submission and is not supported. The master and release versions are vastly improved, supported, and are the recommended way to use this system.

Contributing

My goal with this project is to have a simple LiDAR-Inertial odometry system that can work with minimal friction, in the lines of "it just works". I gladly welcome any contribution or feedback you think would help in this direction. Performance improvements or bug fixes are of course always appreciated.

Thanks to the following contributors

Acknowledgements

KISS-ICP, Kinematic-ICP, Bonxai, PlotJuggler, Rerun

This package is inspired by and would not be possible without the work of KISS-ICP and Kinematic-ICP. Additionally, we use and rely heavily on, either in the package itself or during development, Bonxai, PlotJuggler, Rerun, and of course ROS itself.

A special mention goes out to Rerun for providing an extremely easy-to-use but highly performative visualization system. Without this, I probably would not have made a python interface at all.

License

This project is free software made available under the MIT license. For details, see the LICENSE file.