Guidestar-Free Adaptive Optics with Asymmetric Apertures

Ashok Veeraraghavan; Christopher A. Metzler; Haiyun Guo; Weiyun Jiang

arxiv: 2602.07029 · v3 · submitted 2026-02-02 · 📡 eess.IV · cs.CV

Guidestar-Free Adaptive Optics with Asymmetric Apertures

Weiyun Jiang , Haiyun Guo , Christopher A. Metzler , Ashok Veeraraghavan This is my paper

Pith reviewed 2026-05-16 07:59 UTC · model grok-4.3

classification 📡 eess.IV cs.CV

keywords adaptive opticsphase retrievalmachine learningasymmetric aperturesguidestar-freewavefront sensingaberration correctioncomputational imaging

0 comments

The pith

Asymmetric apertures and machine learning enable the first closed-loop adaptive optics system to correct aberrations in real time without a guidestar or wavefront sensor.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper shows how to build an adaptive optics system that fixes optical aberrations using only ordinary images of natural scenes. An asymmetric aperture in the pupil plane lets phase-retrieval algorithms extract the needed wavefront information from intensity patterns alone. Two machine-learning models then estimate the point-spread function from the scene data and reconstruct the phase errors, which a spatial light modulator corrects optically. The result is real-time, closed-loop operation on dense scenes viewed through unknown obscurants, achieved with far fewer measurements and much less computation than earlier guidestar-free techniques.

Core claim

The authors demonstrate a guidestar-free adaptive optics framework that places an asymmetric aperture at the pupil to enable phase retrieval, uses machine-learning models to recover the PSF and phase aberrations directly from natural-scene intensity measurements, and applies optical correction via a spatial light modulator, experimentally validating real-time performance on scenes through unknown obscurants while using an order of magnitude fewer measurements and three orders of magnitude less computation than prior guidestar-free methods.

What carries the argument

An asymmetric aperture placed at the pupil plane that breaks symmetry to allow unique phase retrieval from single intensity measurements of natural scenes.

If this is right

The framework operates in closed loop on dense natural scenes viewed through unknown obscurants.
It requires an order of magnitude fewer measurements than existing guidestar-free wavefront-shaping methods.
Computation drops by three orders of magnitude relative to prior approaches.
Optical correction occurs in real time using a spatial light modulator.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The approach could extend to astronomical imaging where bright guidestars are unavailable.
It may enable aberration correction in microscopy or biological imaging without introducing artificial reference points.
Testing under dynamic, time-varying turbulence would reveal whether the machine-learning recovery remains stable outside the reported experimental conditions.

Load-bearing premise

Machine learning models trained or operating on natural-scene intensity measurements can reliably recover the point-spread function and phase aberrations induced by unknown obscurants without additional calibration or reference data.

What would settle it

Running the system on a laboratory setup with a known, controlled phase aberration and natural-scene images, then measuring whether image sharpness improves after correction; failure to improve would falsify the claim.

read the original abstract

This work introduces the first closed-loop adaptive optics (AO) system capable of optically correcting aberrations in real-time without a guidestar or a wavefront sensor. Nearly 40 years ago, Cederquist et al. demonstrated that asymmetric apertures enable phase retrieval (PR) algorithms to perform fully computational wavefront sensing, albeit at a high computational cost. More recently, Chimitt et al. extended this approach with machine learning and demonstrated real-time wavefront sensing using only a single (guidestar-based) point-spread-function (PSF) measurement. Inspired by these works, we introduce a guidestar-free AO framework built around asymmetric apertures and machine learning. Our approach combines three key elements: (1) an asymmetric aperture placed at the system's pupil plane that enables PR-based wavefront sensing, (2) a pair of machine learning algorithms that estimate the PSF from natural scene measurements and reconstruct phase aberrations, and (3) a spatial light modulator that performs optical correction. We experimentally validate this framework on dense natural scenes imaged through unknown obscurants. Our method outperforms state-of-the-art guidestar-free wavefront shaping methods, using an order of magnitude fewer measurements and three orders of magnitude less computation.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

This paper shows a working closed-loop guidestar-free AO setup on natural scenes via asymmetric apertures plus ML for PSF and phase estimation, but the experimental support is thin on details.

read the letter

The main point is that they close the loop on adaptive optics without a guidestar or wavefront sensor. They place an asymmetric aperture in the pupil, use one ML model to pull the PSF from intensity images of dense natural scenes, feed that into a second model for phase reconstruction, and apply the correction optically with a spatial light modulator. Experiments claim it handles unknown obscurants and beats prior guidestar-free methods with an order of magnitude fewer measurements and three orders less compute. That integration of the old Cederquist asymmetric-aperture idea with Chimitt-style ML is the concrete step forward, and the hardware-in-the-loop results are the part that could matter for microscopy or remote sensing setups that want simpler optics.

Referee Report

2 major / 1 minor

Summary. The paper claims to introduce the first closed-loop adaptive optics (AO) system for real-time optical correction of aberrations without a guidestar or wavefront sensor. It combines an asymmetric aperture for phase retrieval, two machine-learning models that estimate the PSF from natural-scene intensity images and reconstruct the phase aberrations, and a spatial light modulator for correction. Experimental validation is reported on dense natural scenes imaged through unknown obscurants, with claimed outperformance over prior guidestar-free wavefront-shaping methods using an order of magnitude fewer measurements and three orders of magnitude less computation.

Significance. If substantiated, the result would be significant for AO applications where guidestars are unavailable, such as imaging through atmospheric turbulence or biological media. By extending asymmetric-aperture phase retrieval with ML for efficient, real-time operation on natural scenes, the work addresses a long-standing practical barrier and reports substantial efficiency gains over existing computational approaches.

major comments (2)

[Abstract and Experimental Validation section] Abstract and Experimental Validation section: the claim of quantitative outperformance on natural scenes through unknown obscurants is unsupported by visible error bars, dataset sizes, number of trials, or ablation studies on the ML models; these details are load-bearing for the central assertion that the system reliably recovers PSF and phase without calibration or reference data.
[ML Models section] ML Models section (around the pair of algorithms for PSF estimation and phase reconstruction): the generalization of the learned mapping to truly novel, unseen obscurants and scene statistics is asserted but lacks cross-validation results or domain-shift analysis between training and deployment distributions, which directly affects whether the closed-loop correction is guidestar-free in the strong sense claimed.

minor comments (1)

[Abstract] Clarify the precise definitions of 'order of magnitude fewer measurements' and 'three orders of magnitude less computation' with explicit baseline comparisons and timing breakdowns.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their insightful comments. We address each major comment point-by-point below, indicating where revisions will be made to strengthen the manuscript.

read point-by-point responses

Referee: [Abstract and Experimental Validation section] Abstract and Experimental Validation section: the claim of quantitative outperformance on natural scenes through unknown obscurants is unsupported by visible error bars, dataset sizes, number of trials, or ablation studies on the ML models; these details are load-bearing for the central assertion that the system reliably recovers PSF and phase without calibration or reference data.

Authors: We agree that the quantitative claims would be more robust with additional supporting details. In the revised manuscript, we will include error bars on all reported performance metrics, explicitly state the dataset sizes and number of trials, and add ablation studies on the ML models. These changes will better substantiate the outperformance and the reliability of guidestar-free PSF and phase recovery. revision: yes
Referee: [ML Models section] ML Models section (around the pair of algorithms for PSF estimation and phase reconstruction): the generalization of the learned mapping to truly novel, unseen obscurants and scene statistics is asserted but lacks cross-validation results or domain-shift analysis between training and deployment distributions, which directly affects whether the closed-loop correction is guidestar-free in the strong sense claimed.

Authors: We note that our experiments already employed distinct training and testing sets with different obscurants to evaluate performance under novel conditions. To directly address the request for stronger evidence, the revised manuscript will add k-fold cross-validation results and explicit domain-shift analysis (e.g., distribution distance metrics) between training and deployment data. This will more clearly support generalization in the guidestar-free setting. revision: yes

Circularity Check

0 steps flagged

Minor self-citation to prior ML wavefront sensing work, but no load-bearing circularity

full rationale

The paper's derivation chain consists of an asymmetric-aperture PR setup, two ML models for PSF and phase estimation from natural-scene intensity images, and SLM-based correction, all validated experimentally on dense scenes through unknown obscurants. The only self-citation is to Chimitt et al. for the guidestar-based ML precursor; this is cited only as inspiration and is not used to justify the guidestar-free claim or any equation. No prediction, parameter, or uniqueness result reduces by construction to a fitted input or to the present paper's own outputs. The central claims rest on external experimental measurements rather than self-referential fitting or renaming.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The central claim rests on the empirical performance of two unspecified machine-learning models whose training data, architecture, and loss functions are not described; no free parameters, axioms, or invented entities are explicitly listed in the abstract.

pith-pipeline@v0.9.0 · 5512 in / 1104 out tokens · 38413 ms · 2026-05-16T07:59:17.663383+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

Our approach combines three key elements: (1) an asymmetric aperture... (2) a pair of machine learning algorithms that estimate the PSF from natural scene measurements and reconstruct phase aberrations...
IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

We use the first 36 modes of Zernike polynomials... Phase U-Net is trained... on simulated random phase aberrations

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.