Comparison of Human-Robot Interaction Torque Estimation Methods in a Wrist Rehabilitation Exoskeleton

• Published in: Journal of intelligent & robotic systems Vol. 94; no. 3-4; pp. 565 - 581
• Main Authors: Saadatzi, Mohammadhossein, Long, David C., Celik, Ozkan
• Format: Journal Article
• Language: English
• Published: Dordrecht Springer Netherlands 01.06.2019; Springer; Springer Nature B.V
• Subjects: Accuracy; Algorithms; Artificial Intelligence; Comparative analysis; Control; Disturbance observers; Electrical Engineering; Engineering; Error detection; Evolutionary algorithms; Exoskeletons; Human engineering; Inverse dynamics; Mathematical models; Mathematical optimization; Mechanical Engineering; Mechatronics; Methods; Model accuracy; Parameter estimation; Parameter identification; Parameter sensitivity; Rehabilitation; Robot dynamics; Robotics; Robots; Sensitivity analysis; Sensors; Torque; Wrist; Friction modeling; Sensitivity analysis; System identification; Exoskeletons; Force estimation; Nonlinear disturbance observer
• ISSN: 0921-0296; 1573-0409
• DOI: 10.1007/s10846-018-0786-8

In this paper, experimental implementation and comparative accuracy evaluation of five methods for estimation of human-robot interaction torques are presented. These methods vary from the simplest case of using solely commanded motor torques, to partial consideration of the robot dynamics, to advanced methods considering full robot dynamics such as inverse dynamics (ID) and nonlinear disturbance observer (NDO) based algorithms. Dynamic and friction models of the exoskeleton were developed and their parameters were identified using an evolutionary optimization algorithm to ensure high parameter accuracy. When used with accurate model parameters, ID method led to 20 to 22% average error, while NDO method generated 12 to 18% average error, as evaluated in experiments with a force sensor. These values compare to average error values of up to 132% for using motor torques only, and between 25 to 69% when partial dynamics were used. A sensitivity analysis of the ID and NDO methods to inaccuracies in model parameter estimations revealed considerable sensitivity of these advanced methods to model parameter variations. A summary is provided for the typical estimation accuracy levels that can be expected of these methods and discuss the limitations and considerations that should be taken into account for their use.