Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Adaptive Invariant Extended Kalman Filter for Legged Robot State Estimation

Kyung-Hwan Kim, DongHyun Ahn, Dong-hyun Lee, JuYoung Yoon, Dong Jin Hyun

TL;DR

This work tackles proprioceptive state estimation for legged robots under varying contact conditions and slips by developing an Adaptive Invariant Extended Kalman Filter on the $SE_{2+N}(3)$ group. The method adaptively tunes the foot-model noise covariance $\hat{Q}_{f_i}$ via online covariance matching and uses a Mahalanobis-distance-based slip rejection within a PKM framework, all while relying on a contact-detection algorithm instead of dedicated contact sensors. Key contributions include: (1) a dynamic foot-noise adaptation mechanism with a moving-window innovation covariance, (2) integration of Mahalanobis-based slip rejection into the PKM, and (3) real-robot validation on LeoQuad showing improved RMSE in velocity and orientation during dynamic locomotion. The approach reduces hardware complexity and enhances robustness, enabling more reliable proprioceptive estimation for legged locomotion in challenging environments.

Abstract

State estimation is crucial for legged robots as it directly affects control performance and locomotion stability. In this paper, we propose an Adaptive Invariant Extended Kalman Filter to improve proprioceptive state estimation for legged robots. The proposed method adaptively adjusts the noise level of the contact foot model based on online covariance estimation, leading to improved state estimation under varying contact conditions. It effectively handles small slips that traditional slip rejection fails to address, as overly sensitive slip rejection settings risk causing filter divergence. Our approach employs a contact detection algorithm instead of contact sensors, reducing the reliance on additional hardware. The proposed method is validated through real-world experiments on the quadruped robot LeoQuad, demonstrating enhanced state estimation performance in dynamic locomotion scenarios.

Adaptive Invariant Extended Kalman Filter for Legged Robot State Estimation

TL;DR

This work tackles proprioceptive state estimation for legged robots under varying contact conditions and slips by developing an Adaptive Invariant Extended Kalman Filter on the group. The method adaptively tunes the foot-model noise covariance via online covariance matching and uses a Mahalanobis-distance-based slip rejection within a PKM framework, all while relying on a contact-detection algorithm instead of dedicated contact sensors. Key contributions include: (1) a dynamic foot-noise adaptation mechanism with a moving-window innovation covariance, (2) integration of Mahalanobis-based slip rejection into the PKM, and (3) real-robot validation on LeoQuad showing improved RMSE in velocity and orientation during dynamic locomotion. The approach reduces hardware complexity and enhances robustness, enabling more reliable proprioceptive estimation for legged locomotion in challenging environments.

Abstract

State estimation is crucial for legged robots as it directly affects control performance and locomotion stability. In this paper, we propose an Adaptive Invariant Extended Kalman Filter to improve proprioceptive state estimation for legged robots. The proposed method adaptively adjusts the noise level of the contact foot model based on online covariance estimation, leading to improved state estimation under varying contact conditions. It effectively handles small slips that traditional slip rejection fails to address, as overly sensitive slip rejection settings risk causing filter divergence. Our approach employs a contact detection algorithm instead of contact sensors, reducing the reliance on additional hardware. The proposed method is validated through real-world experiments on the quadruped robot LeoQuad, demonstrating enhanced state estimation performance in dynamic locomotion scenarios.
Paper Structure (11 sections, 32 equations, 4 figures, 1 table)

This paper contains 11 sections, 32 equations, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. Prerequisites
  3. $SE_{2+N}(3)$ Group Representation
  4. Right Invariant Extended Kalman Filter
  5. Methodology
  6. System Dynamics & Measurement Model
  7. Contact Detection Algorithm
  8. Adaptive Filtering - Foot Model Noise Estimation
  9. Adaptive Filtering - Slip Rejection
  10. Experimental Validation
  11. Conclusion

Figures (4)

  • Figure 1: The snapshot of robot, LeoQuad, experiment. The top row is a snapshot of a pronk on flat terrain, while the bottom is a snapshot of a flying trot on rough terrain.
  • Figure 2: Flowchart of Adaptive IEKF algorithm.
  • Figure 3: Graph showing the relationship between foot noise and the Root Mean Square Error(RMSE), illustrating the effect of noise levels on the filter accuracy. The experiments were conducted on rough terrain using a flying trot.
  • Figure 4: Estimated contact, slip, and $\alpha$ values of the left front leg during the flying trot on rough terrain. $\alpha$ is normalize by $\alpha_{max}$ for readability. The gray-shaded regions indicate contact phases.