Constrained canonical transformations used for representing system parameters in an adaptive control of manipulators

In the computed torque control of strongly coupled nonlinear multivariable systems such as robots, the practical lack of the exact dynamic model of the system to be controlled as well as the presence of a dynamic interaction with an unmodeled environment presents considerable difficulty in control. The traditional way of using the Denavit-Hartenberg conventions for constructing the dynamic model and representing the unknown system parameters results in considerable computational complexity. A novel approach to this problem based on canonical transformations as improvements of an initial, very simple system model was recently being investigated from many aspects. Though it was able to produce certain adaptivity, its limitations were also revealed. The aim of this paper is to present a possible improvement of this method resulting in the combination of soft computing and simple numerical algorithms of uniform structure independent of the particular structure of the robot arm to be controlled. The improved method is discussed and its abilities are illustrated by simulation results.