Proprioceptive Estimation of External Wrenches for Legged Robots Using Enhanced Momentum-Based Kalman Filter

Abstract

Fast and accurate measurement of external disturbances is critical for legged robots in complex environments. While physical force/torque (F/T) sensors provide direct measurements, they are susceptible to high-frequency noise and mechanical damage. To address this challenge, this article proposes a proprioceptive measurement method that fuses multisensor data for high-precision external wrench estimation. The core of the method is an enhanced momentum (EM)-based estimator combined with a Kalman filter (KF). The conventional momentum-based method is improved by incorporating a derivative term. The enhanced term reduces the phase lag caused by low-pass filtering methods. To mitigate the noise amplified by the derivative term, an augmented state-space measurement model is established and is used for the KF to achieve accurate estimation. The proposed method relies entirely on proprioceptive feedback and resolves the tradeoff between measurement response and noise immunity. Comparative simulations and hardware experiments on a quadruped robot validate the accuracy and versatility of the proposed method, demonstrating that the EM-based KF can serve as a reliable measurement method for external force estimation without requiring physical force sensors.