RESPLE2: Inertial-Informed Recursive Non-Uniform Spline Estimation for Multi-Sensor Odometry with Global Refinement

YouTube | arXiv | Website | Dataset

RESPLE2 is a recursive non-uniform B-spline state estimation framework supporting a versatile range of multi-sensor odometry configurations, enabling LiDAR-inertial, LiDAR-visual-inertial (LVIO), LiDAR-leg-inertial (LKIO), and LiDAR-leg-visual-inertial (LKVIO) odometry systems with a unified design with optional global spline correction. All runs comfortably within real-time constraints.

BibTex Citation

@article{cao2026resple2,
  title={RESPLE2: Inertial-Informed Recursive Non-Uniform Spline Estimation for Multi-Sensor Odometry with Global Refinement},
  author={Cao, Ziyu and Talbot, William and Li, Song and Chirarattananon, Pakpong and Gustafsson, Fredrik and Hutter, Marco and Li, Kailai},
  journal={arXiv preprint arXiv:2610.37602},
  year={2026}
}

Dependencies

RESPLE2 supports ROS 2 Humble on Ubuntu 22.04 and ROS 2 Jazzy on Ubuntu 24.04. Install the corresponding ROS 2 distribution, then source it and install the dependencies:

# Choose the setup file for your ROS 2 distribution:
source /opt/ros/humble/setup.bash
# source /opt/ros/jazzy/setup.bash

sudo apt install build-essential cmake python3-colcon-common-extensions python3-yaml \
  libpcl-dev libeigen3-dev libopencv-dev libyaml-cpp-dev libboost-thread-dev \
  libsuitesparse-dev libceres-dev
sudo apt install ros-${ROS_DISTRO}-pcl-conversions ros-${ROS_DISTRO}-cv-bridge ros-${ROS_DISTRO}-image-transport \
  ros-${ROS_DISTRO}-rosidl-default-generators ros-${ROS_DISTRO}-rosidl-generator-dds-idl \
  ros-${ROS_DISTRO}-ament-lint-auto ros-${ROS_DISTRO}-ament-lint-common
# Optional: for playing MCAP bags
sudo apt install ros-${ROS_DISTRO}-rosbag2-storage-mcap

Install Pinocchio 3.9.0. Set pinocchio_DIR in CMakeLists.txt to its CMake directory (currently /usr/local/lib/cmake/pinocchio).

Compilation

source /opt/ros/${ROS_DISTRO}/setup.bash
mkdir -p ~/ros2_ws/src
cd ~/ros2_ws/src
git clone --recursive https://github.com/ASIG-X/RESPLE2.git
cd ~/ros2_ws
colcon build --symlink-install resple2
source install/setup.bash

Global refinement also requires rviz2_plugin_ser2res to clear the RViz displays when the map is updated. Clone it into the same workspace's src/ directory, build it with colcon build --symlink-install --packages-select rviz2_plugin_ser2res from the workspace root, and source install/setup.bash before launching. The provided RViz configuration already includes this plugin.

Usage

For LIO use, set if_visual: false and if_kin: false in the YAML file selected by the launch file under resple2/config/. Enable if_visual for visual fusion and if_kin for kinematic fusion when the corresponding sensor data is available.

• GrandTour dataset (LKVIO)

source install/setup.bash
ros2 launch resple2 resple_groundtour.launch.py
# Open another terminal and run:
source install/setup.bash
ros2 bag play /path/to/bag/

• JEPLO dataset (LKIO)

source install/setup.bash
ros2 launch resple2 resple_jeplo.launch.py
# Open another terminal and run:
source install/setup.bash
ros2 bag play /path/to/bag/

• PPS Tracked Vehicle dataset (LVIO)

source install/setup.bash
ros2 launch resple2 resple_ppstv.launch.py
# Open another terminal and run:
source install/setup.bash
ros2 bag play /path/to/bag/

• HOPPER dataset (LIO)

source install/setup.bash
ros2 launch resple2 resple_hopper_ext.launch.py # or: ros2 launch resple2 resple_hopper_int.launch.py
# Open another terminal and run:
source install/setup.bash
ros2 bag play /path/to/bag/

• HANDI dataset (LVIO)

source install/setup.bash
ros2 launch resple2 resple_handi_ar.launch.py # or: ros2 launch resple2 resple_handi_az.launch.py; ros2 launch resple2 resple_handi_mz.launch.py
# Open another terminal and run:
source install/setup.bash
ros2 bag play /path/to/bag/

To enable global refinement, set if_loop_closure: true in the YAML configuration file used by the selected launch file under resple2/config/.

Contributors

Ziyu Cao (Email: ziyu.cao@liu.se)

William Talbot (Email: wtalbot@ethz.ch)

Kailai Li (Email: kailai.li@liu.se)

Credits

Thanks for SFUISE, RESPLE, FAST-LIVO2, R3LIVE, iBTC, Livox-SDK, and basalt.

License

The source code is released under GPLv3 license. For commercial use, please contact Ziyu Cao at ziyu.cao@liu.se or Kailai Li at kailai.li@liu.se to discuss an alternative license.