Legged Robot State Estimation

Multi-Sensor Fusion for Quadruped Robot State Estimation using Invariant Filtering and Smoothing [RA-L'25] :
This letter presents two multi-sensor state estimation frameworks for quadruped robots, based on the Invariant Extended Kalman Filter (InEKF) and Invariant Smoother (IS). These frameworks fuse kinematics, IMU, LiDAR, and GPS data to mitigate position drift—particularly along the z-axis—which is a common issue in proprioceptive-based approaches. Specifically, we derive observation models that satisfy group-affine properties, enabling the integration of LiDAR odometry and GPS into the InEKF and IS frameworks. We evaluate the proposed frameworks against LiDAR-based odometry methods in both indoor and outdoor experiments, achieving lower Relative Position Error (RPE) and significantly reduced Absolute Trajectory Error (ATE).
Invariant Smoother for Legged Robot State Estimation With Dynamic Contact Event Information [T-RO'23] :
This article proposes an invariant smoother for legged robot state estimation using IMU and leg kinematics, assuming static foot contact. By leveraging state-independent Jacobians through group-affine residuals, the method achieves improved convergence, especially during dynamic contact events. It also introduces a slip rejection method and a contact loop model to enhance robustness and accuracy in both indoor and long-distance outdoor experiments.
Legged Robot State Estimation With Dynamic Contact Event Information
​[RA-L'21] :
This letter proposes a state estimation algorithm for legged robots using a MAP formulation solved via the Gauss-Newton method with Schur Complement-based marginalization. The algorithm leverages SO(3) manifold structures and state reparameterization for efficient computation. A slip rejection method is also introduced, and the approach is validated against the IEKF in diverse real-world experiments.

Parameter Identification

Online Friction Coefficient Identification for Legged Robots on Slippery Terrain Using Smoothed Contact Gradients [RA-L'25] :
While the friction coefficient is an important parameter for control and state estimation in legged robots, estimating it in an online manner remains challenging. In particular, due to the nonsmooth nature of contact dynamics, exact gradients often become uninformative for identifying the friction coefficient. In this work, we propose an online friction coefficient identification framework for legged robots operating on slippery terrain. To address the issue of non-informative gradients, we introduce analytically smoothed gradients with respect to the friction coefficient, which demonstrate faster and more consistent performance compared to the baseline in robot experiments.