Simulation of online human arm inertia estimation for robot-aided rehabilitation

Abstract

Most of the studies on rehabilitation robots consider the human arm inertia and the gravity torque as system disturbances. Individual anthropometry varies from patient to patient, and therefore human limbs are not modelled. Some studies used the Disturbance Observer (DOB) as a method of disturbance rejection. However, if the inertia and gravity torque parameters of the human arm could be estimated, they could be effectively used in the controller loop to achieve precise motion control. This paper proposes a novel Reaction Torque Observer (RTOB) based estimation technique which updates parameters using learning and recursive algorithms in real-time. The proposed method is applicable to many robot systems where the load inertia or the load is not known. A simulation was carried out with realistic parameters to compare the performance of two competing methods proposed namely, Adaptive Linear Neuron (ADALINE) and Recursive Least Squares (RLS). Results show that the RLS method outperforms the ADALINE method based on the performance criteria of accuracy, precision and convergence speed for estimating the inertia.