Learning Control in the Presence of Measurement Noise

This paper examines the properties of learning control algorithms for linear MIMO systems in the presence of measurement noise. The only information used about the measurement errors is the magnitude upper bounds on the errors. Several control algorithms that are convergent (that is result in asymptotically perfect tracking) under perfect measurements are considered. It is shown that despite the imperfect measurements the learning controller can improve the tracking performance by driving the tracking error asymptotically within certain bounds which depend on the upper bound on the disturbance and the system parameters. The results help the designer to assess the extend of performance deterioration of learning control systems under influence of measurement errors.