Optical see-through

Optical see-through (OST) is a display approach for augmented reality in which the user views the real world directly through transparent optics, while virtual imagery is superimposed onto that view by a partially reflective element. It contrasts with video see-through, where the real world is captured by cameras, combined with computer graphics electronically, and shown on an opaque screen. In an OST system the light from the surroundings reaches the eye optically rather than as a digitised video stream, so the real scene is seen at the resolution and latency of natural vision.

The technique is the basis of most transparent AR headsets and smart glasses, including the HoloLens, Magic Leap devices, and the Epson Moverio line. It is most often realised in an optical see-through head-mounted display (OST-HMD), and it descends directly from the head-up displays used in aircraft and from Ivan Sutherland's 1968 head-mounted display, which already combined a computer image with a direct view of the room through half-silvered mirrors.

How it works

An optical see-through display places one or more optical combiners in front of the eyes. A combiner is partially transmissive, so the user can look directly through it at the real world, and partially reflective, so the eye also receives virtual images generated by a small display engine (microdisplay) and routed into the combiner. Ronald Azuma's 1997 survey describes the arrangement as similar to a head-up display except that the combiner is attached to the head, and characterises optical see-through HMDs as "a HUD on a head".

Because the combiner behaves like a half-silvered mirror, it admits only part of the incoming real-world light while reflecting part of the display light into the eye. Azuma notes an example device that transmitted roughly 30 percent of the light from the real world, and observes that most optical see-through HMDs reduce real-world light enough to act like a pair of sunglasses even when switched off. The amount of light each path contributes is a design choice; a combiner can also be tuned to reflect particular wavelengths, which suits a single-colour display.

Several optical structures are used to carry the virtual image to the eye:

Combiner type	Principle	Example users
Beam splitter / partial mirror	A flat or curved half-silvered surface reflects the display image toward the eye while transmitting the real scene	Early military and research HMDs
Birdbath optics	A beam splitter paired with a curved mirror folds the light path, allowing a compact micro-OLED display and a relatively wide field of view at lower cost	Xreal glasses
Waveguide (diffractive, reflective, holographic)	Light is injected into a thin transparent substrate, propagates by total internal reflection, and is coupled out across an eyebox; allows a glasses-like form factor	HoloLens 2, Magic Leap 2, Lumus reflective waveguides

Diffractive and reflective waveguides have become the dominant approach for slim transparent headsets and glasses because they decouple the bulky display engine from the lens worn in front of the eye. Lumus, whose reflective geometric waveguides are used in Meta's Ray-Ban Display glasses announced in 2025, states that reflective waveguides use total internal reflection to keep light inside the substrate and so reach higher optical efficiency and transparency than diffractive designs.

Comparison with video see-through

Combining the real and the virtual is the central design decision in any AR system, and the two basic options are optical and video. Rolland, Holloway and Fuchs set out a detailed comparison of the two in 1995, and Azuma's survey enumerates the tradeoffs.

According to Azuma, optical see-through has four main advantages over video:

• Simplicity. There is only one image stream to manage, the graphics, because the real world is seen directly through the combiner with a delay of a few nanoseconds. Video must digitise and synchronise separate real and virtual streams, adding at least one frame of delay.
• Resolution. The real-world view is not degraded by a sensor or a screen, so it is seen at the full resolving power of the eye; video limits both the real and the virtual to the resolution of the display.
• Safety. If power is lost the user still has a direct view of the surroundings (the headset becomes heavy sunglasses), whereas a video see-through user is effectively blind because the display is opaque.
• No eye offset. The eyes stay where they are, so the viewpoint matches natural vision; video puts the user's effective viewpoint at the cameras, which are usually offset from the eyes and may not match the user's interpupillary distance.

The corresponding disadvantages of optical see-through are largely consequences of letting real light through uncontrolled:

• No occlusion. The combiner cannot block light from the real world, so virtual objects appear ghost-like and semi-transparent and cannot hide what is behind them. Azuma calls occlusion one of the strongest depth cues, and its absence is the most cited weakness of the approach; building a combiner that selectively blocks real light is difficult because the blocking filter would have to sit at a point where the image is in focus, which is the eye itself.
• Limited field of view. Wide fields of view are harder to achieve optically, because any distortion in the user's view of the real world has to be corrected with optics rather than in software, and the necessary optics are bulky and expensive.
• Temporal mismatch. The real view is essentially instantaneous while the virtual view is delayed, and the two cannot be matched; a video system can deliberately delay the real-world video to line it up with the rendered image.
• Fewer registration and brightness-matching options. The only information about head position comes from the head tracker, since there is no digitised image of the scene to analyse, and the large dynamic range of a directly viewed real scene makes the brightness of real and virtual objects hard to match. A bright environment washes out the virtual image; a dim one washes out the real.

Because of the flexibility that digital compositing gives video see-through, especially proper occlusion, Azuma suggested it might eventually produce more convincing blends, while optical see-through remained attractive where direct vision, safety and full real-world resolution matter. A 2021 ACM Computing Surveys review of OST-HMDs framed the long-running research goal as "indistinguishable" augmented reality, in which virtual content is hard to tell apart from real objects through the optics.

Persistent technical problems

Two issues beyond occlusion recur in optical see-through design. The first is focus. In an optical system the virtual image is projected at a fixed (or only slowly adjustable) distance, while real objects sit at varying distances, so if the virtual and real distances do not match the user cannot bring both into sharp focus at once. This is closely tied to the vergence-accommodation conflict: most waveguide displays present the image collimated, fixing it near optical infinity, so the eye converges to the virtual object's apparent depth but accommodates to a different one. Research prototypes have used focus-tunable or varifocal optics to supply correct focus cues; a 2017 demonstration by researchers at UNC and collaborators used deformable membrane mirrors to vary focus from about 20 cm to optical infinity across a 100 degree field of view, but such mechanisms have remained too bulky for an eyeglasses form factor.

The second is the additive nature of the display, which cannot render the colour black; black is simply the absence of added light, so dark virtual content becomes transparent. Magic Leap 2, released in 2022, was the first commercial AR headset with dynamic dimming, an electrically controlled layer that lowers the lens transmission from about 22 percent down to roughly 0.3 percent. Its segmented mode can dim only the regions where virtual content appears, improving contrast and giving a partial substitute for true occlusion. Magic Leap's VP of optical engineering Kevin Curtis presented the details at SPIE Photonics West in 2022.

Uses and current status

Optical see-through is used where the wearer needs to keep a real, unmediated view of their surroundings while reference data or 3D graphics are overlaid. Azuma's survey grouped early applications into medicine, manufacturing and repair, visualisation, robot path planning, entertainment and military aviation, and noted that many assembly and repair prototypes chose optical approaches partly for the cost and safety advantages. Modern enterprise and field uses include remote assistance, logistics and surgical guidance, the markets targeted by HoloLens, Magic Leap and the Epson Moverio glasses.

The status of optical see-through hardware shifted notably in the mid-2020s. Microsoft confirmed in October 2024 that it had ended production of the HoloLens 2, with security support promised through the end of 2027, and in February 2025 said it would not develop a HoloLens 3, ending its first-party AR headset line. Magic Leap shifted toward licensing its optics and waveguide manufacturing, and Epson continued to sell its binocular Moverio BT-45C and BT-45CS see-through glasses, which use a Si-OLED engine and a 34 degree field of view, for industrial and remote-support work. At the same time, consumer momentum moved toward lighter optical see-through glasses driven by reflective waveguides and birdbath optics, such as Xreal's display glasses and Meta's Ray-Ban Display, which pairs a Lumus waveguide with an LCOS microdisplay.

This optical approach stands apart from the passthrough-based mixed reality of headsets such as the Apple Vision Pro and Meta Quest 3, which are video see-through: their displays are opaque and the wearer sees the room through outward-facing cameras rather than through transparent optics.