Model predictive control of long short-term memory forecasting disturbances for adaptive optics system

In this study, we present a comprehensive exploration of the integration of deep learning models into the predictive control framework of adaptive optics systems. Specifically, we focus on the utilization of Shack-Hartmann sensors and deformable mirrors for wavefront sensing and correction, and employ Long Short-Term Memory (LSTM) neural networks for environmental modeling. Our results show that this method can not only eliminate aberrations but also achieve tracking of laser focus points, which is highly beneficial for rapid volumetric scanning multiphoton microscopy. We discuss the feasibility, methodology, potential advantages, and challenges of this innovative approach, as well as its implications for advancing the field of adaptive optics.