Fri, Dec 10
Speaker: Sang Hoon Kang - Dr. Zhang's group
Title: Stochastic estimation of human arm impedance with robots having nonlinear joint friction - Degradation of accuracy and reliability of estimation and its practical remedy
Abstract: The basic assumption of stochastic human arm impedance estimation methods is that the human arm and robot behave linearly for small perturbations. In the present work, we have identified the degree of influence of nonlinear friction in robot joints to the stochastic human arm impedance estimation. Internal model based impedance control (IMBIC) is then proposed as a means to make the estimation reliable and accurate by compensating for the nonlinear friction.
From simulations with a nonlinear Lugre friction model, it is observed that the reliability and accuracy of the estimation are severely degraded with nonlinear friction: below 2 Hz, multiple and partial coherence functions are far less than unity; estimated magnitudes and phases are severely deviated from that of a real human arm throughout the frequency range of interest; and the accuracy is not enhanced with an increase of magnitude of the force perturbations. In contrast, the combined use of stochastic estimation and IMBIC provides with accurate estimation results even with large friction: the multiple coherence functions are larger than 0.9 throughout the frequency range of interest and the estimated magnitudes and phases are well matched with that of a real human arm. Furthermore, the performance of suggested method is independent of human arm and robot posture, and human arm impedance.
In order to confirm the enhanced accuracy and reliability of estimation under nonlinear friction, a two-DOF SCARA-type robot with significant nonlinear friction was used for the experiments, with IMBIC and with a proportional and derivative(PD) control (the simplest form of impedance control), respectively. After the stochastic estimation method with the SCARA-type robot and IMBIC was first validated with a spring array, it was applied to the estimation of human arm impedance. The experimental results show that the stochastic estimation with IMBIC yields accurate and reliable estimation even under substantial friction: the multiple coherence functions exceed 0.95 throughout the frequency range of interest and the estimated magnitudes and phases are well matched with a second-order best-fit model.
Furthermore, the best-fit model shows a reasonable agreement with the results of previous research works. Stochastic estimation with IMBIC has shown its effectiveness for the estimation of human arm impedance with conventional robots.
Therefore, the IMBIC will be useful in measuring human arm impedance with conventional robot, as well as in designing a spatial impedance measuring robot, which requires gearing.
This is of practical importance, since it now opens to the way of using existing conventional robots that tends to be easier, cheaper and faster, to implement than designing and prototyping special, dedicated robots.