Predictive control for adaptive optics using neural networks

Adaptive optics (AO) has become an indispensable tool for ground-based telescopes to mitigate atmospheric seeing and obtain high angular resolution observations. Predictive control aims overcome latency in AO systems: the inevitable time delay between wavefront measurement correction. A current method of predictive uses empirical orthogonal functions (EOFs) framework borrowed from weather prediction, but advent modern machine learning rise neural networks (NNs) offer scope further improvement. Here, we evaluate potential application NNs highlight advantages that they offer. We first show their superior regularization over standard truncation used by linear EOF with on-sky data before demonstrating NNs’ capacity model nonlinearities on simulated data. This is highly relevant operation pyramid sensors (PyWFSs), as handling would enable a PyWFS be low modulation deliver extremely sensitive measurements.