Research journey

What we have
found so far

The story of this search, step by step, with the logic drawn out. It ends with the newest condition from the BD team: keep only devices from the last three years, and what that rule really does to the list.

The path

How we got from the brief to today

Each step changed the question, so each step changed the shortlist. Reading them in order shows why the answer moved.

Step 1 · the brief

Three features for the client's demo

The worker on the floor needs three things: share a live view with a remote expert, live translation across Japanese, Thai, and English, and place a 3D object to measure a real part. Target demo: the Germany expo, October 2026.

Step 2 · the scope cut

Remote help reduced to share the view plus voice

The client relaxed the remote-assist part. No 3D annotation, no certified AR platform needed. Any video call that shows the live camera and voice is enough. This reopened the device choice to cheaper and lighter options.

Step 3 · the wider survey

More devices, and a clear rating system

The BD team asked to add RayNeo X3 Pro, XREAL Air 2 Ultra, Rokid AR Lite, Vuzix M400, SiNGRAY G2, and Quest 3, and to score each device with the ◎ ○ △ × system. The SiNGRAY G2 and Quest 3 are headsets, kept as input for the next phase.

Step 4 · two research passes

Verify every claim against the source

Two source-checked research passes read the vendor pages directly. The headline result: no single hands-free device does all three jobs on its own software, so a two-device combination is needed.

Step 5 · a correction fixed

The Meta Ray-Ban, set straight

An earlier note was wrong. The Ray-Ban Meta does share a live view, but today only inside Meta's own apps, never Zoom or Teams. Meta has since opened a developer-preview toolkit that lets a few partners like Microsoft pull the camera into their own app, though it is not general yet and is not a call you join. We verified this against Meta's own pages and corrected the record. Its translation now covers Japanese, Thai, and English online, so the gap is the camera path, not the languages.

Step 6 · measurement, we build it answered

We will build the 3D measurement ourselves

New decision: develop the measurement in-house, aiming for rough sizing of about 1 to 2 cm, glass-only if we can, or glass plus phone. The research has landed. At that accuracy you do not need LiDAR. The safe path is the phone measuring while the glass shows the result. Glass-only is realistic only on a 6DoF camera device like the Quest 3. Full answer is below.

Step 7 · the recency rule this page

Keep only the last three years

The BD team suggests excluding older devices and limiting the list to models released within the last three years. It is a sensible instinct, because this technology moves fast. Below is what the rule does, and the small change that makes it safe.

The new condition, drawn out

The three-year rule, on a timeline

Every device placed by its model launch year. The shaded zone is the last three years, 2023 to 2026. Color shows status today: in the window, older but still sold and supported, discontinued, or upcoming.

Inside the window (2023+) Older, but still sold and supported Discontinued or end of life Upcoming
Industrial assisted-reality Consumer AI glasses AR / spatial glasses MR / VR headset

Years are approximate model-launch years. Status is 2026 sales and support. The dashed line marks the three-year cutoff.

The catch: a strict rule deletes the family that does the main job

Look at the left of the chart. The whole industrial assisted-reality family sits there: Vuzix M400 (2019), Epson Moverio (2021), Moziware Cimo and RealWear Navigator 520 (both 2022). A literal three-year cut throws every one of them out.

But that is the family that natively does see-what-I-see with your own Zoom, Teams, or VSight. These are older models, yet Vuzix, RealWear, and Moziware still sell and support them in 2026. Cutting them by launch date alone would remove the proven workhorses and leave mostly consumer glasses that cannot run your support tool.

The fix: judge by "still sold and supported," not by launch year

Keep the three-year idea as the default, it correctly drops dead tech. But measure it by active sales and support in 2026, not the launch date. That keeps a still-supported workhorse like the Vuzix M400 in the running, and still removes the genuinely dead devices regardless of age.

Dropped as end of life: Microsoft HoloLens 2 (2019, winding down), Magic Leap 2 (2022, discontinued), and even a newer one, the XREAL Air 2 Ultra (2024) is already discontinued. Age is not the real signal. Support is.

A reply you can send

Answering the BD team directly

Draft reply

Thank you. I agree we should drop older technology, and a three-year window is a good default. We do not need a longer timeframe to catch new devices.

One small adjustment. I suggest we measure recency by whether a model is still sold and supported in 2026, not only by its launch year. The reason is that the industrial glasses that do the core remote-help job, like the Vuzix M400, launched before 2023 but are still actively sold and updated. A strict launch-date rule would remove exactly the devices that fit the task best.

So the rule I propose: keep anything still on sale and supported, and drop anything discontinued or end of life even if it is newer. That removes HoloLens 2 and Magic Leap 2, and also the XREAL Air 2 Ultra, while keeping the proven workhorses. I have put the full timeline and logic on our shared page.

The newest finding

Building the measurement ourselves

You decided to develop the 3D measurement in-house, glass-only if we can, or glass plus phone. The research landed. Good news first: at ±1–2 cm you do not need LiDAR. The deciding factor is which device gives open access to its camera and sensors, so we can build on it.

Phone plus glass safe path

The phone does the measuring, the glass shows the result. Proven and buildable today.

  • The phone runs Google's ARCore Depth API. It builds depth from a normal moving camera, with no special depth sensor, and is most accurate from 0.5 to 5 m, which covers a 1 to 2 m working distance.
  • Works with the industrial glasses you wear (Vuzix, RealWear, Moziware) as the display. They are 3DoF with no depth, so they cannot measure on their own.
  • Build on ARCore for Android or ARKit for iPhone. Mature, documented, low risk.

Glass-only stretch goal

Measurement runs on the worn device, no phone. Only a 6DoF device with open camera access can do it.

  • Meta Quest 3 / 3S is the strongest, and verified. Its Passthrough Camera API gives raw camera frames with calibration, plus a Depth API and a room Mesh. The full kit, ready to publish. It is a headset though.
  • Apple Vision Pro can too, but raw camera needs a special Apple enterprise licence that expires and is in-house only.
  • XREAL 6DoF glasses (Light) expose the camera, 6DoF, and a depth mesh through NRSDK. The Air 2 Ultra has only grayscale tracking cameras (no usable RGB camera) and is discontinued.

To build measurement on the glass itself, a device needs two things at once: open access to its camera and sensors, and real 6DoF plus depth sensing. Drawn out, it is clear why only a few devices land in the buildable corner, and why the industrial glasses you wear need a phone.

Glass-only buildable open camera + 6DoF + depth Wear it — a phone measures No usable camera locked open Developer access to the camera and sensors → On-device sensing: 3DoF → 6DoF + depth Meta Quest 3 Apple Vision Pro (enterprise) XREAL Aura (upcoming) XREAL Air 2 Ultra (to confirm) Vuzix M400 Moziware Cimo RealWear 520 Ray-Ban Meta Even Realities G2
Industrial assisted-reality Consumer AI glasses AR / spatial glasses MR / VR headset

A phone running ARCore or ARKit would sit in the top-right buildable corner too. That is exactly why the phone-plus-glass path works: the phone brings the corner, the glass brings the eyes.

Honest accuracy note

No vendor document promises ±1–2 cm at 1 to 2 m. The number is realistic for a single, near-range shot that casts a ray onto a detected surface, but it must be tested on the real device, not trusted from the spec sheet. Avoid plain anchor-to-anchor SLAM distance, its drift ranged from 1 to 43 cm in tests. Use the depth and surface APIs instead.

Dead-ends for a self-built measure app

Meta Ray-Ban gives no camera to outside code. Even Realities G2 has no camera at all. RealWear and the Vuzix Z100 SDKs expose only voice and display, not the sensors. These can still be the display end of a phone-plus-glass design, but they cannot host the measurement.

Where we stand

The shortlist after the recency rule

Devices that pass the refined rule (still sold and supported) and matter for the worn role, scored with the BD team's system. The build-measure column now reflects the verified in-house build research. Click any device for the full detail.

excellent yes partial / to confirm× no
DeviceRecencyShare viewJP + TH + ENBuild measure

The build-measure column shows whether the device can host our own ±1–2 cm app. means glass-only is verified (only the Quest 3 so far). means phone-assist only, the glass shows the result while a phone running ARCore or ARKit does the measuring. × means it cannot host it. See the Building section above, and the strategy and explorer for device detail. Device photos are from each maker's site, used for identification (Quest 3 and Ray-Ban Meta via Wikimedia, CC). Open the device gallery to read each one's full story.

The journey continues

Two answers landed this week: the corrected Meta facts, and how we can build the measurement. Next we validate the ±1–2 cm target on real hardware, and confirm the camera and SDK access on the newer devices still inside the three-year window. This page updates as each result lands.