Pedestrian Dead Reckoning using Invariant Extended Kalman Filter

TL;DR

This paper introduces a cost-effective inertial pedestrian dead reckoning method using an invariant extended Kalman filter, demonstrating improved robustness and ease of tuning in GPS-denied environments for bipedal robots.

Contribution

It develops a Lie group-based InEKF for pedestrian dead reckoning and validates its effectiveness through multiple real-world experiments.

Findings

InEKF outperforms standard EKF in accuracy and robustness.

InEKF is easier to tune than EKF.

Method is feasible for real-world bipedal robot navigation.

Abstract

This paper presents a cost-effective inertial pedestrian dead reckoning method for the bipedal robot in the GPS-denied environment. Each time when the inertial measurement unit (IMU) is on the stance foot, a stationary pseudo-measurement can be executed to provide innovation to the IMU measurement based prediction. The matrix Lie group based theoretical development of the adopted invariant extended Kalman filter (InEKF) is set forth for tutorial purpose. Three experiments are conducted to compare between InEKF and standard EKF, including motion capture benchmark experiment, large-scale multi-floor walking experiment, and bipedal robot experiment, as an effort to show our method's feasibility in real-world robot system. In addition, a sensitivity analysis is included to show that InEKF is much easier to tune than EKF.