On-line Parameter Estimation For Robot Manipulators

摘要

In the area of on-line parameter estimation, it is recognized that the performance of standard least square method is unsatisfactory due to its lack of exponential convergence and its incapability of estimating time-varying parameters. Exponential forgetting ofthe past data has been widely suggested to overcome these difficulties, but how the variable forgetting factor should be chosen has not been satisfactorily resolved. In this paper, we propose a novel way of generating variable forgetting factor, namely, varying the forgetting factor according to the norm of the gain matrix, which leads to an easily analyzable estimator with exponential parameter convergence and bounded gain. We study this variable-forgetting estimator in the context of robotic load estimation for convenience although the result is generally applicable for any linearly- parameterized estimation model, We also describe the application of the result to improve the performance of our recent composite and indirect adaptive controllers.