AI Wavefront Estimation for the Rubin Observatory Active Optics System

The Vera C. Rubin Observatory will use an Active Optics System (AOS) to correct for alignment and mirror surface perturbations introduced by gravity and temperature gradients during survey commissioning and operations. This system uses out-of-focus images from wavefront sensors located at the edge of the Rubin focal plane to learn and correct for perturbations in the optical system. We show the results of a deep learning approach for wavefront sensing that has been integrated within the Rubin AOS and compare these results to current techniques being used to estimate the Rubin wavefront. We evaluate the speed of the deep learning approach and its robustness to the effects of vignetting at the edge of the focal plane, blending of out-of-focus images, errors in the centroiding of the images, and signal-to-noise. We discuss the advantages of this approach including its simplicity and the potential to directly relate the out-of-focus images to the optical state of the system as well as some of the challenges associated with training such a system.