Interactive model identification for nonholonomic cart pushed by a mobile manipulator

A model identification method for unknown environments has been developed. By the interactions between a mobile manipulator and the unknown object, a nonholonomic cart, sensory information has been collected to estimate the model parameters of the cart, which are used to control the cart. Since the raw data are contaminated by noise they can not be modeled statistically, a wavelet based least square method is proposed to estimate these parameters for the cart. The raw signal has been decomposed into a certain bandwidth to generate a series of new signals, which are used to estimate the parameters. The new signal, which has the minimal estimation residual in least square sense is adopted as the best estimation. The error convergence of the estimation approach is given. The experimental results indicate that the estimation accuracy can be significantly improved by the use of the proposed method.