Estimation of Human Compliance Parameters During Robot Kinesthetic Teaching by Compensation of the Dynamics of the Force-Torque Sensor

This paper presents a method for estimating the compliance of the human arm in a robot kinesthetic teaching scenario, while taking into account the dynamics of the force-torque sensor. We model the dynamics of the sensor as a second order system and identify this experimentally. With the estimated transfer function, the sensor dynamics can be removed from the measurements of the human-robot interaction experiment, such that only the human dynamics is measured. Once the signal has been filtered, we use regularized Dynamic Regressor Extension and Mixing to estimate these parameters with better convergence characteristics than previously-used estimators. We validate our results both in simulation and experimentally on a real robot.