Functional learning in optimal non-linear control

It is demonstrated that least squares functional learning can be successfully applied to the control of nonlinear plants when augmented with a plant linearization, in the case of stochastic or functional disturbances, and with unmodeled dynamics. However, as expected, the adaptive-Q scheme is more responsive to effects that are not state dependent. The learning-Q method with the interpolation functions on a 4 by 4 grid robustly controls the plant, but as expected, is surpassed by an adaptive scheme with learning enhancements. Work is currently being undertaken to improve the algorithm by means of truncation of the adaptation to those interpolation functions closest to the current region of interest to allow much finer grid spacings leading to further improvements in the control.