Augmented Vision

Scope. The underlying display and projection technologies that put information in front of (or into) the eye — light engines, combiners, and projection paths — as distinct from the product history and sensing stacks covered by smart-glasses and project-aria. These are the techniques behind the current smart-glasses wave: waveguide optics, birdbath combiners, direct retinal projection, adaptive-optics correction, and the light-engine choices (LCoS, microLED, laser beam scanning) that determine brightness, colour, and form factor. The same delivery problem — coupling a controlled light field into a moving pupil — recurs throughout retinal-prosthesis engineering and the patent collection in visual-prosthesis-patents.

The system stack

Every architecture solves the same chain: pixels → light engine → relay optics → combiner or projection path → pupil → retina. The free parameters are where the image is formed (microdisplay, waveguide, mirror, or directly on the retina), how the virtual light is combined with the real world while remaining see-through, and how bright, efficient, and compact the chain can be made. The targets are set by human vision: field of view (FOV), eyebox (the pupil-position volume over which the image is visible), angular resolution, dynamic range, correct depth cues, and ergonomics — with strong trade-offs between all of them. ¹

A recurring human-factors problem is the vergence–accommodation conflict (VAC) — the displayed image sits at a fixed focal depth while vergence cues say otherwise; the review notes VAC can be worse in AR than VR because virtual content is superimposed on a correctly-focused real world. Adaptive-focus displays and retinal projection are two proposed answers to this. ¹

Concrete anchor points from shipping or prototype hardware: Meta Ray-Ban Display puts ~1 lumen into its optics and ~5,000 nits at the eye, with an effective resolution of ~400×400 (stated 600×600), 40 pixels/° at 20° diagonal; Even G2 delivers 640×350 at 27.5° with 1,200 nits and 98% see-through — the whole field is still far below foveal acuity, which is why foveated and non-uniform projection schemes keep appearing in patents. ^{2 3}

Delivery optics: how light reaches the eye

Freeform prism and the first birdbath — Google Glass

Google Glass (2013) used an LCoS microdisplay (Himax, field-sequential LED illumination) coupled through polarizing beam splitters to a solid prism combiner — structurally, the second type of birdbath (fully reflective curved mirror; the user looks through the beam splitter only), with the projector placed to the side because the FOV was tiny (640×360). Discontinued in 2015 after the privacy backlash; enterprise editions survived until 2023. ^{4 5}

Birdbath optics — Xreal, Lenovo, Rokid, Viture

The common consumer “big screen” glasses optic. Light from an OLED microdisplay is polarized, hits a polarizing beam splitter, and is reflected to a curved partially-reflective mirror (60/40 transmission in the analysed Nreal design); a quarter-wave plate rotates polarization 90° so the return image passes through the PBS to the eye. Measured numbers: only ~15% of display nits reach the eye and ~26% of real-world light gets through (~75% blocked, “moderately dark sunglasses”); the whole stack is ~25 mm thick and front-heavy; a front polarizer blocks ~95% of forward-projected light (patients’ viewing privacy) and additional coatings suppress mirror-like reflections. Strengths: cheap, good image quality with Lambertian OLEDs, works with mid-range FOV (~50°). ⁵

Diffractive waveguides — the dominant AR path

The image is in-coupled by a diffractive grating into a thin transparent slab, propagates by total internal reflection, and is out-coupled by gratings toward the eye; intermediate grating regions expand the pupil (exit pupil expansion, EPE). The concept traces to Nokia Research — Tapani Levola’s exit-pupil-expander patents (filed 2004) and his 2006 SID paper, building on 1990s holographic-combiner work — so today’s AR optics industry is an old idea that took 15 years of materials science to manufacture. Diffractive elements come in two families: holographic optical elements (wavefront-recorded) and lithographically-made surface-relief gratings (SRGs). ¹

Key engineering variables:

• Substrate refractive index sets FOV (via étendue): ordinary glass ~1.8; lithium niobate ~2.3; silicon carbide ~2.7 — the highest known for optics. Meta’s Orion had to develop optical-grade SiC (normally an opaque green electronic material), diamond-tooled processing, and the industry’s first slant-etched gratings; one plate replaced a three-plate glass stack that showed “rainbows everywhere,” and the material’s thermal conductivity also helps. Cost and wafer scarcity (millions of dollars of R&D; EV-era supply chains now over-built and dropping prices) are the open question. ~70° FOV demonstrated in glasses form factor. ⁶
• Colour dispersion forces one waveguide per colour in RGB systems (e.g., three for Magic Leap 2, which chose LCoS over microLED and laser-beam scanning after evaluating all three; ~2,000 nits at 70° with per-colour LED illumination routed through the waveguide to avoid a beamsplitter). ⁷
• Artifacts: rainbows/glints from ambient light, ghost images, and eye glow (forward-projected display light). Diffractive glow is typically 50–100% of the light the user sees; graded gratings can spread brightness more evenly (used in Even G2). ^{2 8}

Geometric (reflective) waveguides — Lumus

A stack of embedded partially-reflecting mirrors (“slats”) inside glass replaces the gratings: 3–7× more efficient than diffractive waveguides at equal FOV/eyebox, with better colour uniformity and a small fraction of the eye glow (measured ~1.5% on Meta Ray-Ban Display vs 50–100% typical for diffractive). Historically limited by manufacturability/yield (aligned mirror stacks), which is why military/medical/industrial users adopted it first; Meta’s first display product (Ray-Ban Display, 2025) uses a Lumus Z-Lens geometric waveguide with an OmniVision LCoS microdisplay and a Goertek-designed engine — a striking “shipping beats R&D” choice given Meta’s in-house SiC/microLED research programme. Lumus was founded by Yaakov Amitai, whose holographic-optics lineage reaches back to 1990s holographic combiner patents. ²

Holographic combiners — Apple’s approach

Apple’s head-mounted optics programme (inventor Richard J. Topliss and colleagues) has patented a foveated, hologram-based architecture quite distinct from the waveguide monoculture. US20190285897A1 (“Display device”, Apple Inc, filed 2017) describes a reflective holographic combiner recorded with point-to-point holograms: a laser-diode-array projector feeds a distribution waveguide, MEMS scanning mirrors write through layered pupil-expanding waveguides, and separate foveal and peripheral projectors write wide or narrow beams onto matching eye-box points to “match the visual acuity of the eye” — four foveal + four peripheral projectors in one claim, with gaze tracking selecting which to drive. US12,681,308 (granted Jul 2026) adds multi-layer holographic combiners whose second layer focuses each replicated image onto the eye box and whose hologram optical power varies across the eye box to compensate aberrations, with an image sensor for distortion-feedback. Topliss’s portfolio also includes a whole tunable-lens family (deformable, fluid-filled, SMA- and voice-coil-actuated lenses) — the makings of an all-optical corrective AR stack. ^{9 10 11}

Foveated dual-module optics — Meta

Meta’s granted counterpart to Apple’s foveated-combiner filings is the foveated display system family (US11294184B2, 2022; EP4088155B1, granted 2026): a high-resolution foveal module — display, MEMS beam-steering mirror, focusing stage, eye tracking, and an angular/wavelength-selective holographic lens — writes a virtual image into the foveal region of the tracked pupil, while a peripheral module throws low-resolution wide-FOV imagery through a cholesteric-liquid-crystal diffuser embedded inside a pancake lens; real-world light passes through both stages. A switchable Pancharatnam–Berry phase grating stack (US11668932B2) multiplies the steering range so the foveal image can cover the whole field of view, and the diffuser-in-pancake construction is separately claimed (US11448803B1). The four January-2020 filings, now granted across US and EP, are the industry’s first granted block on hardware-foveated near-eye optics — the optical analogue of foveated rendering, and the direct foil to uniform-pixel waveguide displays (see visual-prosthesis-patents).

Direct retinal projection — no combiner at all

Scanned laser beams are rastered through the pupil to write the image directly on the retina: no screen, no combiner, no intermediate image, and a large depth of focus (the eye’s accommodation is taken out of the focus equation). Apple’s live patent family (US11157072B1 — see visual-prosthesis-patents for the full mechanism — plus continuation US12,093,446B1, granted Sep 2024) explicitly targets the accommodation–convergence conflict, with a bright-pupil IR gaze-tracking loop (quadrant-cell PSD) that steers the scanned field into the moving pupil and an azimuth-synchronised adjustable focusing lens. Intel’s Vaunt (2018) demonstrated the concept in ordinary-looking glasses with a VCSEL in the temple, then cancelled it within months. The commercial survivor is medical: QD Laser’s RETISSA / VISIRIUM (Japan) is a laser-retinal-projection low-vision aid (“projects images onto the retina like a planetarium creates a starry sky”), designed within Class 1 laser-safety limits and reported to outperform conventional low-vision aids on distance and near vision. Challenges for a consumer revival: pupil steering robustness, speckle, colour, and laser safety. Meta explored the same idea from the display side — WO2023219925A1 (2023, since lapsed) steers a collimated beam pixel-by-pixel with a micromirror array toward the tracked pupil centre (see visual-prosthesis-patents). ^{12 13}

A further miniaturisation path — displays inside contact lenses — stalled when Mojo Vision ceased smart-lens work in 2023 and pivoted to microLED display fabrication; the corrective-contact-lens branch lives on in azalea-vision. ¹⁴

Light engines

The pixel source determines brightness and colour capability:

Engine	Principle	Used in	Strengths	Limits
LCoS (liquid crystal on silicon)	reflective panel, field-sequential LED light	Google Glass (Himax); Magic Leap 1/2; Meta Ray-Ban Display (OmniVision)	mature, full colour, high resolution; industry default for waveguide AR	needs illumination optics; efficiency moderate
MicroLED	emissive GaN-on-silicon	Even G1/G2 (green, Jade Bird Display); many 2024–26 AR designs; Meta Orion prototypes	very bright, small, efficient	full colour and yield still maturing; Meta chose LCoS for its shipping product
Laser beam scanning (MEMS-LBS)	RGB lasers + MEMS mirrors raster the beam	Apple retinal projectors; Microsoft HoloLens 2; Intel Vaunt	tiny, high contrast, focus-free	speckle, colour registration, eye safety
OLED microdisplay	emissive	birdbath device class (Xreal et al.)	excellent image quality in birdbaths	limited brightness (~800 nits typical)
DMD (digital micromirror device)	MEMS mirror array modulating an external source per pixel (TI DLP); tens of thousands of mirrors tilt between ‘on’ (into the projection pupil) and ‘off’ (beam dump)	DLP projectors; vision-science patterned-stimulus rigs for retinal & prosthesis studies	high-resolution, fast binary patterns; wavelength-agnostic (VIS–NIR lasers/LEDs); polarisation-independent	needs a light source + projection optics; not emissive; the chip must be imaged onto the retina

^{2 5 3}

DMDs as patterned-stimulus engines. Outside consumer displays, the digital micromirror device is the workhorse of patterned light stimulation in vision research. Arens-Arad et al. (2016) built a head-mounted DMD projector for freely-moving rats around a Texas Instruments LightCrafter 3000 DLP platform (684×608 mirrors, 10.8 μm pitch): the stock red LEDs were replaced by a 525 nm LED for natural retinal stimulation and a 915 nm diode laser for activating photovoltaic subretinal implants in sighted animals without also driving healthy photoreceptors, with a custom lens periscope relaying the pattern directly onto the retina, simultaneous visual-cortex recording (VEPs / single units) and full ZEMAX optical characterisation (MTF >0.8). Nine years later the same toolchain appears in prosthetic characterisation: the 2025 tellurium-nanowire nanoprosthesis study integrated a DMD with a multielectrode array and a 635 nm laser, projecting circle/triangle patterns onto the nanowire film (42.5 mW/mm² over a 1,200×1,200 μm area) for spatially resolved recordings. ^{15 16}

How it relates to the other engines here. Micromirror arrays appear twice on this page: as the DMD, switching all mirrors in parallel to compose a whole pattern, and in Meta’s WO2023219925A1 above (see visual-prosthesis-patents), where an addressable micromirror array steers one beam per object point to the pupil. Like LCoS, the DMD modulates a separate light source rather than emitting — but its binary mechanical tilt is polarisation-independent and wavelength-agnostic, exactly what patterned stimulation at 525/635/915 nm needs. Functionally it is the research twin of the direct retinal projector above: both write an image straight onto the retina on purpose — there for AR, here for controlled visual or prosthetic stimulation in animals (see retinal-prosthesis). ^{15 16}

The clinical flagship — PRIMA. The same DMD principle runs the clinical PRIMA retinal prosthesis (Pixium Vision → science-corp): its glasses capture the scene with a camera, process it in a pocket unit, and project the pattern onto the implanted chip with pulsed 880-nm near-infrared light. The engine is a pulse-width-modulated DMD — each micromirror switching in sync with the light-source pulses and duty-cycle-limited for ocular/implant safety (Pixium patent US12061332) — and the 2 × 2 mm, 378-pixel photovoltaic implant under the retina converts the light pattern directly into charge-balanced biphasic currents stimulating bipolar cells: the projected DMD pattern is the stimulation, site by site. Operating point: 30 Hz frame rate (above flicker fusion), percept brightness set by pulse duration (0.7–9.8 ms at 3.5 mW/mm² peak) rather than intensity. First generation was opaque VR glasses (PRIMA-1: 17.5° field, ~10.5 µm optical resolution); since October 2019 they are transparent AR (PRIMA-2: 18.5°, 6.7 µm), projection adapted to the patient’s refraction, allowing simultaneous prosthetic central and natural peripheral vision, with electronic zoom and an optional tinted lens for bright environments.^{17 18 19} And the research loop is closing: Science Corp’s next-generation filings (US20250249280) put eye tracking behind the DMD — pupil/corneal features steer the beam coarsely, retinal features (cells, vasculature, the implant itself) fine-correct the pattern (see retinal-prosthesis and visual-prosthesis-patents).

Adaptive optics and the vision-correction layer

Consumer AR glasses are converging with prescription eyewear, and this is where “adaptive optics” enters in a looser sense than astronomy: not deformable mirrors, but the co-design of display optics with the wearer’s refraction.

• Even HAO (Holistic Adaptive Optics) — Even Realities’ branded approach: projector, waveguide, and lens “tuned as one orchestra,” with a 40%-smaller brighter projector, credit-card-thin multi-layer lenses (centre 30% slimmer), graded output gratings for display uniformity, digitally-surfaced free-form prescription lenses (ray-traced, diamond-turned), UltraFit optical alignment (tilt/vertex/wrap), and 100+ layer anti-reflective/anti-eye-glow coatings — resulting in 98% see-through and a binocular green microLED display at 640×350, 27.5°, 1200 nits. ^{8 3}
• Tunable / autofocus spectacles for presbyopia: liquid-crystal and pixelated-liquid-crystal lenses (IXI, which steers focus using eye movements; Morrow e-Progressives; DeepOptics 32°N; Elcyo). The same lens-actuation technologies appear in Apple’s HMD patents (deformable and fluid-filled tunable lenses) — display-neutral “optics that adapt to the eye” is a shared programme. ¹¹
• Clinical reuse of the same toolbox: RETISSA (retinal projection for low vision, above) and raytrx (OcuLenz AR glasses that remap the visual field around AMD scotomas) show consumer AR display technology flowing directly into vision aids. ¹³

Product map

Device (year)	Delivery optics	Light engine	Key display specs	Status
Google Glass (2013)	prism combiner (birdbath type 2)	LCoS (Himax)	640×360, monocular	discontinued 2015; enterprise to 2023
Microsoft HoloLens 2 (2019)	diffractive waveguide	MEMS laser beam scanning	~40% see-through	enterprise
Magic Leap 2 (2022)	3× high-index glass waveguides	LCoS, per-colour LED	70° diag, ~2,000 nits	enterprise
Even G1 (2024) / G2 (2026)	graded diffractive waveguide	green microLED (JBD in G1)	G2: 640×350, 27.5°, 1,200 nits, 98% see-through, binocular	consumer
Meta Ray-Ban Display (2025)	Lumus geometric waveguide	OmniVision LCoS (Goertek engine)	600×600 (~400×400 effective), 40 px/°, 20° diag, ~5,000 nits, monocular	consumer
Meta Orion (2024)	SiC diffractive waveguide	microLED	~70° FOV	prototype
Xreal One / One Pro (2024–25)	birdbath / flat-prism TIR variant	OLED	~50° FOV	consumer
Apple (2016–2026 filings)	direct retinal projector; multi-layer holographic combiners	laser / MEMS scanning	—	patents; family live (2024 grant)
QD Laser RETISSA (Japan)	retinal projection	laser scanning (VISIRIUM)	low-vision medical aid	clinical / commercial (JP)
PRIMA glasses (Science Corp, 2017–2026)	DMD projector → subretinal photovoltaic chip (no combiner)	pulsed DMD, 880 nm	18.5° field, 6.7 µm optical resolution, 30 Hz	clinical — CE mark Jul 2026

^{4 2 6 3 18}

Open questions

• MicroLED vs LCoS for full colour: the industry’s central fork. Even ships green microLED; Meta and Magic Leap still ship LCoS in 2025–26 despite microLED research programmes.
• Silicon carbide supply chain: if substrate costs keep falling (EV overcapacity), SiC waveguides could become the standard; otherwise it remains a flagship-only material.
• Will retinal projection reach consumers? Apple’s family is live and Intel’s attempt died young; Meta’s 2022-filed micromirror-projection application has already lapsed — eye safety and pupil-steering robustness remain the watch items.
• Foveated, eye-box-matched projection (Apple’s holographic-combiner claims) — a departure from uniform-pixel waveguide orthodoxy; if realised, it would align display sampling with retinal sampling for the first time in consumer hardware.
• Prescription convergence: with Even HAO-style co-design and Apple’s tunable-lens family, the boundary between “AR glasses” and “adaptive spectacles” may dissolve.

See Also

• smart-glasses — products, history, and the sensing stack this page complements
• project-aria — Meta’s machine-perception research glasses (the sensing half)
• visual-prosthesis-patents — patent collection including Apple’s direct retinal projector (US11157072B1)
• retinal-prosthesis — therapeutic light delivery to the retina (PRIMA DMD glasses & chip); same optical problem set
• raytrx — AR glasses for AMD (OcuLenz) — consumer optics as vision aid
• azalea-vision — smart contact lens for optical correction
• digital-therapeutics-for-vergence — display hardware applied to binocular rehabilitation

References

• Xiong, J., Hsiang, E.-L., He, Z., et al. (2021). Augmented reality and virtual reality displays: emerging technologies and future perspectives. Light: Science & Applications 10, 216. ¹
• Guttag, K. (KGOnTech): Meta Ray-Ban Display Part 1 — Lumus waveguide, OmniVision LCoS, Goertek engine (2025-10-30); Nreal Teardown Part 1 — Clones and Birdbath Basics (2021-06-01); Magic Leap 2 at SPIE AR/VR/MR 2022 (link); Hololens 2 display evaluation Part 4 — LBS optics (2020, link); Even Realities G1 (2024, link).
• Meta (2025-03-06). Crystal Clear: Our Silicon Carbide Waveguides & the Path to Orion’s Large FoV. ⁶
• Even Realities: Even G2 technology (Even HAO) and Even G2 product specifications. ^{8 3}
• Apple patent filings: US11157072B1 (granted 2021); US12,093,446B1 (granted 2024); US20190285897A1 (application); US12,681,308 (granted 2026); portfolio at Justia. ¹¹
• Meta display patents: EP4088155B1 & US11294184B2 (foveated display system); US11448803B1 (pancake lens with diffuser); US11668932B2 (switchable PBP grating stack); WO2023219925A1 (virtual reality display system; ceased). See visual-prosthesis-patents.
• QD Laser — VISIRIUM Technology / RETISSA retinal-projection eyewear. ¹³ Clinical performance: Review of Optometry, “Head-Mounted Laser Projector May Outperform Other Low Vision Aids” (link).
• Google Glass — Wikipedia. ⁴
• Nokia/Levola exit-pupil-expander patent US7764413B2 (filed 2004); Levola, T. (2006). Diffractive optics for virtual reality displays. J. SID 14, 467–475. (links)
• Intel Vaunt — The Verge (2018-02-05, link); cancelled April 2018.
• Mojo Vision contact-lens programme ceased, microLED pivot — The Verge (2023-01-07, link).
• Autofocus spectacles — CNN (IXI, Elcyo; link); MAFO (Morrow, DeepOptics; link).
• Arens-Arad, T., Farah, N., Ben-Yaish, S., Zlotnik, A., Zalevsky, Z., & Mandel, Y. (2016). Head mounted DMD based projection system for natural and prosthetic visual stimulation in freely moving rats. Scientific Reports 6, 34873. ¹⁵
• Wang, S., Jiang, C., Yu, Y., et al. (2025). Tellurium nanowire retinal nanoprosthesis improves vision in models of blindness. Science 388, eadu2987 — patterned-stimulus platform: DMD + 635 nm laser + multielectrode array. ^{16 20}
• Palanker Lab (Stanford). Photovoltaic Retinal Prosthesis for Restoration of Sight in Retinal Degeneration — system design page (DMD projection of 880-nm patterns). ¹⁹
• Palanker D, Le Mer Y, Mohand-Said S, et al. Simultaneous perception of prosthetic and natural vision in AMD patients. Nat Commun 2022;13:513. ¹⁷
• Muqit MMK, Le Mer Y, Olmos de Koo L, et al. Prosthetic visual acuity with the PRIMA system at 4 years follow-up. Ophthalmol Sci 2024. ¹⁸

Footnotes

1. raw/articles/xiong-2021-ar-vr-displays-review.md ↩ ↩² ↩³ ↩⁴

2. raw/articles/meta-ray-ban-display-teardown-kguttag-2025.md ↩ ↩² ↩³ ↩⁴ ↩⁵

3. raw/articles/even-realities-g2-product-page-2026.md ↩ ↩² ↩³ ↩⁴ ↩⁵

4. raw/articles/google-glass-wikipedia-2026.md ↩ ↩² ↩³

5. raw/articles/nreal-teardown-birdbath-basics-kguttag-2021.md ↩ ↩² ↩³

6. raw/articles/meta-orion-silicon-carbide-waveguides-2025.md ↩ ↩² ↩³

7. kguttag.com Magic Leap 2 at SPIE AR/VR/MR 2022 ↩

8. raw/articles/even-realities-tech-hao-2026.md ↩ ↩² ↩³

9. raw/articles/apple-us20190285897-holographic-combiner-2019.md ↩

10. raw/articles/apple-us12681308-holographic-combiners-2026.md ↩

11. raw/articles/topliss-apple-hmd-patents-justia-2026.md ↩ ↩² ↩³

12. raw/articles/apple-us12093446-direct-retinal-projector-2024.md ↩

13. raw/articles/qd-laser-retissa-visirium-2026.md ↩ ↩² ↩³

14. theverge.com Mojo Vision pivot, 2023 ↩

15. raw/papers/arens-arad-2016-dmd-head-mounted-projection.md ↩ ↩² ↩³

16. raw/papers/wang-2025-dmd-platform-excerpt.md ↩ ↩² ↩³

17. raw/papers/palanker-2022-simultaneous-prosthetic-natural-vision.md ↩ ↩²

18. raw/papers/muqit-2024-prima-4-year-followup.md ↩ ↩² ↩³

19. raw/articles/palanker-lab-prima-system-2026.md ↩ ↩²

20. raw/papers/wang-2025-tellurium-nanowire-retinal-nanoprosthesis.md ↩