Accurate ground reaction force (GRF) estimation enhances the adaptability of legged robots in real-world environments. This paper presents a simultaneous GRF and state estimation framework that systematically addresses sensor noise and state-dynamics coupling. A decentralized Moving Horizon Estimation (MHE) approach is proposed, fusing proprioceptive and exteroceptive sensors with deterministic contact constraints in a convex optimization. The method achieves real-time (200 Hz) performance on various legged platforms, including the humanoid robot Bucky, and the quadruped Unitree Go1 & B1.
For more details, please refer to the paper. [arxiv] [youtube]
ROS2 (tested with foxy/galactic)
Go to this [repo] and follow build instruction.
Change this line to your own python site-packages path
Change this line to your own ORB_SLAM3 path
For a detailed list of available interfaces and their usage, please visit the [repo].
FROST is used to generate kinematics libraries for the project.
mkdir ~/ros2_ws
cd ~/ros2_ws
git clone https://github.com/well-robotics/Decentralized_EKF_MHE.git
colcon build --cmake-args -DCMAKE_BUILD_TYPE=Release --symlink-installsource unitree_ros2.sh
ros2 launch go1_example go1_new_launch.pyCompile with Pinocchio for general applicability.