Tempestas ex machina - A review of machine learning methods for wavefront control

Aug 2, 2026

Tempestas ex machina reviews wavefront control methods for ground-based adaptive optics systems, focusing on the extreme adaptive optics context needed for direct imaging. The work explains why classic integral controllers struggle with temporal lag, vibrations, and dynamic errors that are not captured in wavefront-sensor residuals, limiting high-contrast imaging with complex coronagraphs. It surveys how modern machine learning—including linear approaches and newer techniques—can improve control without demanding higher detector speed or sensitivity, and summarizes literature on novel ML strategies such as predictive and state-estimation methods.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages15

Reading Time38 min

AccessFree to download

Introduction

Adaptive optics system overview

Limitations of integral control and need for ML

Machine learning approaches for wavefront control

arXiv :2309 .00730v1 [ astro-ph .EP] 1 Sep 2023 Tempestas ex machina: A review of machine learning methods for wavefront control J. Fowlera and Rico Landmanb a Department of Astronomy & Astrophysics, University of California, Santa Cruz, CA, USA b Leiden Observatory, Leiden University, Leiden, The Netherlands ABSTRACT As we look to the next generation of adaptive optics systems, now is the time to develop and explore the technologies that will allow us to image rocky Earth-like planets; wavefront control algorithms are not only a crucial component of these systems, but can benefit our adaptive optics systems without requiring increased detector speed and sensitivity or more effective and efficient deformable mirrors. To date, most observatories run the workhorse of their wavefront control as a classic integral controller, which estimates a correction from wavefront sensor residuals, and attempts to apply that correction as fast as possible in closed-loop. An integrator of this nature fails to address temporal lag errors that evolve over scales faster than the correction time, as well as vibrations or dynamic errors within the system that are not encapsulated in the wavefront sensor residuals; these errors impact high contrast imaging systems with complex coronagraphs. With the rise in popularity of machine learning, many are investigating applying modern machine learning methods to wavefront control. Furthermore, many linear implementations of machine learning methods (under varying aliases) have been in development for wavefront control for the last 30-odd years. With this work we define machine learning in its simplest terms, explore the most common machine learning methods applied in the context of this problem, and present a review of the literature concerning novel machine learning approaches to wavefront control. Keywords: machine learning, wavefront control, adaptive optics, predictive control, linear quadratic Gaussian control (LQG), neural networks, reinforcement learning, Kalman filtering, empirical orthogonal functions (EOF) 1. INTRODUCTION In the search for and characterization of extrasolar planets, direct imaging is a powerful tool to unlock the detection of planets not available via other detection methods (i.e., planets that will not geometrically transit) and characterize them without having to disentangle the light of the planet from its host star. From the ground, direct imaging is only possible with extreme adaptive optics (exAO) where a very high level of correction over a small field of view enables high contrast imaging. Only stellar point spread functions (PSF) with extremely high Strehl ratios will enable coronagraphy to effectively null host starlight, and result in high enough contrasts to resolve the light of substellar companions. The scope of this review is limited specifically to wavefront control methods used within the field of ground-based adaptive optics for astronomical instrumentation, with a keen interest in this extreme adaptive optics context. (For a review of machine learning methods for wavefront sensing, see Wong, 2021 .1 ) An adaptive optics system at its simplest is made of 3 fundamental components: a wavefront sensor (which records information on the state of the wavefront after it has been aberrated by Earth-atmosphere), a deformable mirror (which attempts to correct for those aberrations), and a control algorithm that determines what correction the deformable mirror (DM) should apply given the wavefront sensor information available. Figure 1 shows a control diagram of this system. To date, every extreme AO system runs their standard control for mid and high order modes as an integrator. This refers specifically to a closed-loop integral style controller, where the wavefront sensor is downstream of the Further author information: (Send correspondence to J.F.) J.F.: [E-mail: jumfowle@ucsc.edu](E-mail: jumfowle@ucsc.edu) Figure 1 . The full state of turbule

FAQs

Why do classic integral controllers in extreme adaptive optics face performance limits?

They do not adequately address temporal lag errors evolving faster than the correction time, nor vibrations or dynamic errors not represented in the wavefront-sensor residuals. These issues degrade high-contrast imaging performance with complex coronagraphs.

What problem motivates applying machine learning to wavefront control?

Future instruments require higher-fidelity correction for detecting and characterizing Earth-like planets, which current control approaches have not achieved. Scaling to larger systems also makes DM-to-sensor calibration increasingly time-consuming, motivating automation via ML.

What ML methods does the review consider for wavefront control?

It surveys common ML methods used in this setting and reviews the literature on novel ML approaches. The keywords indicate techniques such as neural networks, reinforcement learning, Kalman filtering, and linear-quadratic-Gaussian control (LQG).

Aug 2, 2026

Upload by Lucas Martin