BIEVR-LIO: Robust LiDAR-Inertial Odometry through Bump-Image-Enhanced Voxel Maps

BIEVR-LIO is a robust LiDAR-Inertial Odometry framework that uses a high-resolution, voxel-wise oriented height image map to exploit subtle geometric variations in challenging, information-sparse environments.

Abstract

Reliable odometry is essential for mobile robots as they increasingly enter more challenging environments, which often contain little information to constrain point cloud registration, resulting in degraded LiDAR–Inertial Odometry (LIO) accuracy or even divergence. To address this, we present BIEVR-LIO, a novel approach designed specifically to exploit subtle variations in the available geometry for improved robustness. We propose a high-resolution map representation that stores surfaces as voxel-wise oriented height images. This representation can directly be used for registration without the calculation of intermediate geometric primitives while still supporting efficient updates. Since informative geometry is often sparsely distributed in the environment, we further propose a map-informed point sampling strategy to focus registration on geometrically informative regions, improving robustness in uninformative environments while reducing computational cost compared to global high-resolution sampling. Experiments across multiple sensors, platforms, and environments demonstrate state-of-the-art performance in well-constrained scenes and substantial improvements in challenging scenarios where baseline methods diverge. Additionally, we demonstrate that the fine-grained geometry captured by BIEVR-LIO can be used for downstream tasks such as elevation mapping for robot locomotion.

Setup

The core estimator (bievr_lio) is a self-contained, ROS-independent library. On top of it we provide both a ROS1 interface (bievr_lio_ros) and a ROS2 interface (bievr_lio_ros2), which live side by side under interfaces/.

Installation

Dependencies

BIEVR-LIO is intentionally light on dependencies: the core estimator only needs Eigen and Ceres.

Build instructions for both ROS versions are below. Each also offers an optional Docker image for quickly trying out the system without setting up dependencies.

ROS1

For quick testing: Docker

If you just want to try the system out without setting up dependencies, build the image and drop into a shell inside it:

cd docker/
./run_docker_ros1.sh -b

The -b flag builds the image. On subsequent runs you can omit it to reuse the existing image. Your ~/data folder is mounted to /home/bievr/data inside the container so you can keep datasets outside the image.

To open another terminal inside the running container (e.g. to launch a node and play a bag):

docker exec -it BIEVR-LIO-ROS1 /bin/bash

Build

Requires ROS Noetic and python3-catkin-tools (sudo apt install python3-catkin-tools).

Create a catkin workspace and clone BIEVR-LIO into it:

mkdir -p ~/catkin_ws/src
cd ~/catkin_ws
catkin init
catkin config --extend /opt/ros/noetic
catkin config --cmake-args -DCMAKE_BUILD_TYPE=Release
catkin config --merge-devel

cd ~/catkin_ws/src
git clone git@github.com:ethz-asl/BIEVR-LIO.git BIEVR-LIO

Install the Ceres version used by BIEVR-LIO with the provided script (builds Ceres 2.2.0 from source):

./BIEVR-LIO/docker/scripts/install_ceres.sh

(Optional) Livox support. The Livox CustomMsg branches are only compiled if the corresponding driver is found in the workspace at build time. Otherwise BIEVR-LIO builds fine without them. If you need to process Livox data, clone and build the matching driver into ~/catkin_ws/src before building BIEVR-LIO (each driver also needs its Livox-SDK installed system-wide):

• Livox gen1 (livox_ros_driver, enables BIEVR_WITH_LIVOX): livox_ros_driver + Livox-SDK
• Livox gen2 (livox_ros_driver2, enables BIEVR_WITH_LIVOX2): livox_ros_driver2 + Livox-SDK2

Build and source it:

cd ~/catkin_ws
catkin build bievr_lio_ros
source devel/setup.bash

ROS2

For quick testing: Docker

If you just want to try the system out without setting up dependencies, build the image and drop into a shell inside it:

cd docker/
./run_docker_ros2.sh -b

The -b flag builds the image. On subsequent runs you can omit it to reuse the existing image. Your ~/data folder is mounted to /home/bievr/data inside the container.

To open another terminal inside the running container (e.g. to launch a node and play a bag):

docker exec -it BIEVR-LIO-ROS2 /bin/bash

Build

Requires ROS2 Jazzy and python3-colcon-common-extensions (sudo apt install python3-colcon-common-extensions). The system was tested on Jazzy, but other ROS2 distributions might also work.

Create a colcon workspace and clone BIEVR-LIO into it:

mkdir -p ~/colcon_ws/src
cd ~/colcon_ws/src
git clone git@github.com:ethz-asl/BIEVR-LIO.git BIEVR-LIO

Install the Ceres version used by BIEVR-LIO with the provided script (builds Ceres 2.2.0 from source):

./BIEVR-LIO/docker/scripts/install_ceres.sh

(Optional) Livox support. The Livox CustomMsg branch is only compiled if livox_ros_driver2 is found in the workspace at build time. Otherwise BIEVR-LIO builds fine without it. If you need to process Livox data, clone and build the driver into ~/colcon_ws/src before building BIEVR-LIO (it also needs its Livox-SDK2 installed system-wide). Only gen2 exists for ROS2 (enables BIEVR_WITH_LIVOX):

• livox_ros_driver2 + Livox-SDK2

Build and source it (from the workspace root, so colcon picks up both BIEVR/, the core, and interfaces/ros2):

cd ~/colcon_ws
source /opt/ros/jazzy/setup.bash
colcon build --packages-up-to bievr_lio_ros2
source install/setup.bash

Run data

BIEVR-LIO provides two entry points, available for both ROS versions:

• process_topics runs online: it subscribes to the LiDAR and IMU topics and processes messages as they arrive. Use it with a live sensor or alongside rosbag play.
• process_bag reads a recorded bag directly and pushes its messages through the pipeline as fast as they can be processed (no real-time playback). This is the preferred choice for offline evaluation and reproducing results.

In the commands below, replace <sensor_config> with one of the provided configs (see Configuration) or your own. Add rviz:=true to bring up the visualization.

ROS1

Process live topics:

roslaunch bievr_lio_ros process_topics.launch sensor_config:=<sensor_config>

Replay a rosbag:

roslaunch bievr_lio_ros process_bag.launch sensor_config:=<sensor_config> rosbag:=/path/to/bag.bag

ROS2

Process live topics:

ros2 launch bievr_lio_ros2 process_topics.launch.py sensor_config:=<sensor_config>

Replay a rosbag2 directory:

ros2 launch bievr_lio_ros2 process_bag.launch.py sensor_config:=<sensor_config> rosbag:=/path/to/bag_dir

Configuration

The configuration is split in two files:

• config/params.yaml: Algorithm parameters (map resolution, sampling, optimization, IMU window, ...). These are dataset-independent and typically do not need to be adjusted: the defaults have been validated across a wide range of sensors, platforms, and environments.
• config/sensor_configs/<name>.yaml: Per-dataset / per-sensor settings: the LiDAR and IMU topic names, the LiDAR→IMU extrinsic calibration, and the LiDAR min/max range.

Select a sensor config at launch with sensor_config:=<name>, which resolves to config/sensor_configs/<name>.yaml (an absolute path starting with / is used verbatim, so configs may also live outside the package). Likewise params:=<name> (default params) selects config/<name>.yaml.

Provided datasets

We provide ready-to-use sensor configs for the following public datasets:

Config	Dataset
enwide	ENWIDE
ncd	Newer College Dataset
gamma	GEODE
mars	MARS-LVIG
grandtour	GrandTour

Running on your own data

To run BIEVR-LIO on a new sensor or dataset, copy one of the provided sensor configs to config/sensor_configs/<your_name>.yaml and adjust:

• topics.pointcloud / topics.imu : The topic names in your data.
• calibration : the T_IMU_LIDAR extrinsic (LiDAR → IMU) rotation and translation for your setup.
• lidar.min_range_m / lidar.max_range_m : the usable range of your LiDAR.

The algorithm parameters in params.yaml can usually be left at their defaults.

Acknowledgements

We thank the authors of DLIO, Wavemap and UGPM for open-sourcing their works that served as an inspiration for us. We used ascii-image-converter for our ascii art.

Citation

Please cite our work if you are using BIEVR-LIO in your research.

@article{pfreundschuh2026bievr,
title        = {BIEVR-LIO: Robust LiDAR-Inertial Odometry through Bump-Image-Enhanced Voxel Maps},
author       = {Pfreundschuh, Patrick and Tuna, Turcan and {Le Gentil}, Cedric and Siegwart, Roland and Cadena, Cesar and Oleynikova, Helen},
year         = 2026,
journal      = {Robotics: Science and Systems},
}