Inverse Dynamics Controllers for Robust Control: Consequences for Neurocontrollers

It is proposed that controllers that approximate the inverse dynamics of the controlled plant can be used for on-line compensation of changes in the plant's dynamics. The idea is to use the very same controller in two modes at the same time: both for static and dynamic feedback. Implications for the learning of neurocontrollers are discussed. The proposed control mode relaxes the demand of precision and as a consequence, controllers that utilise direct associative learning by means of local function approximators may become more tractable in higher dimensional spaces.