On-sky demonstration of self-learning predictive control with MagAO-X

Abstract

Direct imaging of exoplanets is very tricky and requires extremely well corrected wavefronts. Especially low-order order modes are detrimental to the performance of coronagraphs at their inner-working angle. However, that is precisely where conventional AO systems have the highest residuals that are caused by servo-lag errors. This servo-lag error can be reduced with predictive control where the control anticipates the future state of the atmospheric disturbance. We use a self-learning model predictive controller based on the concepts from sub-space predictive control (SPC). We present a novel implementation of the SPC by using an online QR-decomposition based recursive least squares approach. This approach has now been used for self-learning control of vibrations on the MagAO-X instrument. We see on average an Strehl increase of 15 percent and a decrease of the jitter to 0.9 mas. I will discuss how we have implemented the controller and its on-sky perfomance.