In this guide
How Optical See-Through AR Glasses Work: Beam-Splitters, Head Tracking, and the Phone Link
Optical see-through AR glasses put digital text in front of your eye without taking the world away. Unlike a closed headset that replaces your view with screens, these glasses use a half-reflective film: part of the display light bounces into your eye, and most of the room light passes straight through. The result is a small heads-up display floating over reality — a clock, a notification, a compass heading — while you still see the room, the road, and the people around you.
Why this topic is in the news
On 28 October 2026, XREAL launched Aura at Google's Android Developer Summit as the first wired XR glasses powered by Android XR. According to XREAL's own release, Aura uses optical see-through lenses, Sony Micro-OLED displays at 1200p and 120 Hz, a 70-degree field of view, a split-compute puck, electrochromic dimming, and a reported weight under 95 grams. Those are commercial figures for a finished product, reported here only as context — the rest of this guide explains the physics underneath, and how a student build reaches the same idea with far simpler parts.
How optical see-through works
The core trick is a beam-splitter: a thin film that is partly mirror and partly window. A small display sits above the lens, angled so its light hits the film at about 45 degrees. The film reflects that light sideways into the eye, while room light coming from the front passes through the film and the clear lens to the same eye. Your eye therefore receives two images at once: the world, and the display.
Three consequences follow, and every student build has to live with them:
- Brightness is a contest. The display only wins if it is brighter than the background behind the text. That is why small OLED displays look crisp indoors and wash out in direct sunlight: the sun is simply a stronger light source than the display. Commercial glasses fight this with far brighter micro-displays and with dimming that darkens the background.
- The image floats at a fixed distance. A simple beam-splitter does not focus the display at different depths. The text appears at one comfortable reading distance, and your eye focuses there while the world behind it may be at another distance. This is fine for short text, and it is why these glasses show pages of text rather than objects that sit convincingly on a table.
- One eye or two changes everything. A monocular build serves one eye, so there is no stereoscopic depth. A binocular product serves both eyes with two displays, which allows depth but doubles the optics, the alignment work, and the cost. For a first build, monocular is the honest choice.
How head tracking works
A heads-up display becomes interesting when it reacts to your head. The standard student sensor is the MPU6050, which combines a 3-axis accelerometer and a 3-axis gyroscope on one chip, talking over I2C. The two halves answer different questions:
- The accelerometer feels gravity. When your head is still, the accelerometer reports which way is down, which gives pitch (nodding) and roll (tilting sideways) directly. It cannot tell you your compass heading, because gravity points down no matter which way you face.
- The gyroscope feels rotation. It reports how fast you are turning around each axis. To get heading (yaw), the firmware adds up those rotation rates over time — integration. This works, but small measurement errors accumulate, so the heading slowly drifts away from truth. Without a magnetometer to anchor absolute north, drift is unavoidable, and the practical fix is a re-zero button: hold your head level, press, and the reference heading resets.
Together these give 3 degrees of freedom: orientation only. Position — where you are in the room — is not measured at all, which is why a student build shows head-locked pages that move with you, never windows pinned to a spot in the room. Pinning windows to the room needs positional tracking: cameras, depth sensors, or external beacons, plus mapping software. That is a different project, and naming it as future scope is more honest than pretending orientation alone can do it.
| Sensor part | What it measures | Datasheet range |
|---|---|---|
| MPU6050 accelerometer | Acceleration on 3 axes, including gravity | ±2g to ±16g, selectable |
| MPU6050 gyroscope | Rotation rate on 3 axes | ±250 to ±2000 degrees per second, selectable |
| SSD1306 OLED | The display itself, not a sensor | 128 by 64 pixels over I2C |
Why the phone does the talking
A small microcontroller is a poor place for speech recognition: transcription models need tens of megabytes of memory, while an ESP32-class chip has hundreds of kilobytes of RAM. The architecture that actually works splits the job:
- The glasses capture the trigger — a button press — and handle display and sensing.
- The phone transcribes the spoken command with its own speech service and sends back a short text reply.
- The glasses display that reply over Bluetooth Low Energy, which is built for exactly this kind of traffic: small packets, infrequent, low power.
Notifications flow the same way in reverse: the phone already receives them, so forwarding the sender and a short message line to the glasses is cheap. The design rule is to keep everything crossing the radio short — a line of text, a heading in degrees, a page-change command — and to let the phone own anything heavy.
Why dimming matters
Aura dims its lenses electronically: electrochromic material darkens when voltage is applied, so the wearer can darken the world behind a video and brighten it again for walking around. A student build cannot easily buy or drive electrochromic glass, so the honest analogue is a sliding tint filter clipped over the lens and moved by hand. The physics lesson is identical — contrast is display brightness divided by background brightness, so darkening the background makes dim text readable — and the mechanism is something a student can build, explain, and demonstrate in a viva.
Building it at student scale
If you build the companion project to this guide, the work breaks down into four jobs. First, print the frame and align the optics: the display angle and the film position decide whether the text is readable, so expect several adjustment rounds before anything is glued down. Second, write the firmware: sensor reading, orientation math, page rendering, and button handling on the ESP32. Third, build the companion app: notification forwarding, a voice-command path, and a live yaw readout that proves the link works. Fourth, calibrate: define the level-head zeroing ritual, observe how far yaw drifts after two minutes of wear, and write both facts into the report.
Demonstrate indoors, where the small display wins the brightness contest. Take regular breaks during wear, keep brightness comfortable, and never wear the prototype while driving or cycling. Treat the lithium battery like any other lab cell: correct polarity, a protected holder, and supervision.
See the companion build that puts all of this together: XREAL Aura-Inspired Optical See-Through AR Glasses with Head-Tracked HUD and Voice Assistant — a monocular beam-splitter build with head-tracked compass, phone notifications, voice replies, and a manual tint slider.
FAQ
Why does the display need a beam-splitter instead of a tiny screen in front of the eye?
Because a screen in front of the eye blocks the world. A beam-splitter reflects the display into the eye while transmitting room light, so the wearer sees both at once. That combination — reflection plus transmission in one film — is the whole definition of optical see-through.
Why does the compass heading drift?
Because heading comes from integrating gyroscope rotation over time, and small errors accumulate with no magnetometer to correct them. The fix is procedural, not electronic: re-zero the heading with a level-head button press before each demonstration, and report the drift you observe rather than hiding it.
Can a student build show windows pinned to the room?
Not with orientation sensing alone. Pinned windows need positional tracking and room mapping, which a 3-axis sensor cannot provide. A student build shows head-locked pages that move with the wearer, and treats world-locked windows as future scope.
Does this kind of project need internet?
No. The glasses-to-phone link runs over Bluetooth Low Energy, and notifications and HUD pages work without internet. Voice transcription happens on the phone and follows whatever speech service that phone uses, but the glasses link itself needs no connection.
Why is the display unreadable outdoors?
Because contrast is a contest between display brightness and background brightness, and sunlight is far brighter than a small OLED. The build is demonstrated indoors, where the display wins, and the limitation is stated in the report rather than discovered during the demo.
Limitations
This guide explains principles, not a finished product. A student beam-splitter build is monocular, head-locked, and indoor-only; it has no positional tracking, no hand tracking, and no certified eye-safety rating. Commercial figures quoted here belong to XREAL's Aura launch and are reported references only. If you build the companion project, its page documents the same limits with the same honesty — including the drift you will measure yourself.
Want this built for you, with documentation and viva support? The companion project page linked above carries the full breakdown and the quotation form: XREAL Aura-Inspired Optical See-Through AR Glasses with Head-Tracked HUD and Voice Assistant.