Closed loop predictive control of adaptive optics systems with convolutional neural networks

ABSTRACT Predictive wavefront control is an important and rapidly developing field of adaptive optics (AO). Through the prediction future effects, inherent AO system servo-lag caused by measurement, computation, application correction can be significantly mitigated. This lag impact final delivered science image, including reduced strehl contrast, inhibits our ability to reliably use faint guide stars. We summarize here a novel method for training deep neural networks predictive based on adversarial prior. Unlike previous methods in literature, which have shown results previously generated data or open-loop systems, we demonstrate network’s performance simulated closed loop. Our models are able both reduce effects induced push end reliable with natural stars, improving K-band Strehl compared classical over 55 per cent 16th magnitude stars 8-m telescope. further show that LSTM approaches may better suited high-contrast scenarios where error most pronounced, while traditional feed forward high noise scenarios. Finally, discuss strategies implementing real-time astronomical telescope systems.