Variable-order stochastic adaptive control of robotic manipulators with a flexible forearm

Abstract

An application of recursive covariance lattice filter to the adaptive estimation and stochastic control of a robotic manipulator with one flexible link is presented. Not only the effective order, but also the corresponding parameters of an autoregression moving average with a bias (ARMAB) prediction model of the manipulator are updated by a set of pure order recursive lattice algorithms. The reduced-order prediction model that represents significant dynamics of the plant is used to generate optimal control sequences by minimizing the expectation of a weighted cost functional. In the simulations, the manipulator is modeled by the finite element method and Lagrange's equations. Since the system is highly nonlinear in the large motion, and has nonminimum phase phenomenon as observed in small linear displacements, an additional inner PD controller is implemented to augment the system so as to improve its robustness. The performance and robustness of the variable order stochastic adaptive controller are demonstrated by numerical results.