Reduced filtered dynamic model for joint friction estimation of walking bipeds

The present paper presents a novel method for estimating the joint friction of walking bipeds. It combines a measurement-based method with an adaptive model-based method to estimate the joint friction. The former is used when the feet is in contact, while the latter is used when the leg is swinging. The measurements are the feet forces and the readings of an inertial measure unit located at the biped body. The measurement-based method utilizes these measurements into a reduced filtered dynamic model of the biped to obtain an online estimated filtered version of the joint friction. Once the foot swings, a friction model is adopted to represent the joint friction behavior. The model parameters are adaptively identified using the online estimated filtered friction whenever the foot is in contact. The results are validated using full-dynamics of 12-DOF biped model and show a dramatic tracking of the estimated friction to the true one.