MUSE: A Real-Time Multi-Sensor State Estimator for Quadruped Robots

Ylenia Nisticò, João Carlos Virgolino Soares, Lorenzo Amatucci, Geoff Fink and Claudio Semini

This paper has been accepted to IEEE Robotics and Automation Letters, and it is available at https://arxiv.org/abs/2503.12101

Branch note (unitree_sdk): this branch is developed for full compatibility with the Unitree SDK and ROS2 communication. The code has been tested on a Unitree Go2 robot in both simulation and real-world experiments.

💻 Overview

The muse package provides a ROS node and utilities for estimating the state of a quadruped robot using sensor data. It includes algorithms for state estimation, sensor fusion, and filtering.

This version of the code provides a proprioceptive state estimator for quadruped robots. The necessary inputs are

• imu measurements
• joint states
• force exerted on the feet

To simplify setup, we provide a ready-to-use Conda environment so you do not have to manually resolve dependencies.

Suggestions

Simulation

If you want a running simulation, follow the instructions in basic-locomotion-dls-isaaclab.

Real world experiments

To run this code with Unitree robots, you need to port the URDF of your robot in this folder and connect to the robot through unitree_ros2_dls.

For real-world experiments, we recommend the following repositories to control your robot:

• basic-locomotion-dls-isaaclab
• Quadruped-PyMPC

🦖 Prerequisites

• Conda
• Eigen
• Pinocchio
• vcstool

🛠️ Building and Running

To install and run muse with Conda + ROS2:

1. Clone and create the environment:

   git clone https://github.com/iit-DLSLab/muse.git -b unitree_sdk
   cd muse
   conda env create -f environment.yml
   conda activate muse-ros2

2. Fetch external dependencies (e.g. Point-LIO):

   cd muse_ws
   vcs import src < muse.repos

3. Build with colcon:

   cd muse_ws
   colcon build --symlink-install
   source install/setup.bash

4. Launch the state estimator package:

   ros2 launch state_estimator state_estimator.launch.py

   Or launch the full stack with Point-LIO (see Point-LIO integration):

   ros2 launch muse_point_lio muse_with_point_lio.launch.py

To change the name of the topics, check the config folder.

To visualize your data, you can use PlotJuggler:

ros2 run plotjuggler plotjuggler

---

📡 LiDAR / Point-LIO Integration

MUSE includes a muse_point_lio wrapper package that integrates Point-LIO ROS2 as the exteroceptive odometry source for the MultiSensorFusion plugin.

Architecture

[Point-LIO node]  →  /point_lio/odometry
        ↓
[odom_bridge node]  →  /lidar_odometry     (normalised interface)
        ↓
[MultiSensorFusion plugin]  →  /muse/multi_sensor_fusion
        ↓
[TfStatePublisher plugin]  →  TF: world → base

Go2-specific setup

A ready-to-use Point-LIO config for the Go2 is provided at config/go2_muse.yaml in the muse-integration branch of the forked Point-LIO repo. It sets:

• Input topics: /utlidar/cloud and /utlidar/imu (Unitree SDK2 native driver)
• Extrinsics derived from the Go2 URDF (radar_joint → imu_joint)
• Frame IDs: odom_header_frame_id: odom, odom_child_frame_id: imu

The Point-LIO fork is fetched automatically via muse.repos (see build instructions above).

Selecting a different LiDAR

Pass point_lio_launch_file to switch to any other Point-LIO launch file:

ros2 launch muse_point_lio muse_with_point_lio.launch.py \
  point_lio_launch_file:=mapping_velody16.launch.py

---

⚠️ In this repo we provide an example with the Go2 robot. If you want to test MUSE with another one, you need to add the URDF of your robot in this folder, and (possibly) change the name of the legs in the config files

📜 TODO list

• Extend the code to include exteroception (Point-LIO integration via muse_point_lio)
• Conda-based environment
• Support for ROS2 (on going)

🤗 Contributing

Contributions to this repository are welcome.

Citing the paper

If you like this work and would like to cite it (thanks):

@ARTICLE{10933515,
  author={Nisticò, Ylenia and Soares, João Carlos Virgolino and Amatucci, Lorenzo and Fink, Geoff and Semini, Claudio},
  journal={IEEE Robotics and Automation Letters}, 
  title={MUSE: A Real-Time Multi-Sensor State Estimator for Quadruped Robots}, 
  year={2025},
  volume={10},
  number={5},
  pages={4620-4627},
  keywords={Robots;Sensors;Robot sensing systems;Legged locomotion;Odometry;Cameras;Laser radar;Robot vision systems;Robot kinematics;Quadrupedal robots;State estimation;localization;sensor fusion;quadruped robots},
  doi={10.1109/LRA.2025.3553047}}

Maintainer

This repo is maintained by Ylenia Nisticò