Add a 60 second reach check before any headset lesson

The finding this rests on

Forty adults pointed at dots on a table wearing the same Meta Quest 3. Half saw a drawn world. Half saw the real room through the headset cameras. The two modes bent their aim in opposite directions from the very first block.

The camera passthrough group reached too far, by 0.43 cm. The virtual reality group reached too short, by 0.31 cm. Distance made it worse in virtual reality, adding a further 0.19 cm short for every extra 5 cm of reach. In passthrough that figure was 0.07 cm. The full article is at VR and passthrough bend your reach in opposite directions.

The effect did not stop at the headset. After it came off, virtual reality users started 0.51 cm short and were still about 0.35 cm short 10 to 15 minutes later. Passthrough users drifted back to their own baseline.

We score this paper 5 out of 10. The experiment itself is good. It has a no-headset baseline, two groups, 39 people analysed, 320 pokes each and motion capture at 250 times a second. But it is a preprint with no code, from one lab, on one headset. The authors also say many things differed between the two setups at once, so the study shows that the modes differ and not why. Therefore this is a bet on the effect being real, not on the exact millimetres.

Who has this problem

Picture a training studio of eleven people in a city with a car plant. They build headset courses for factories. Wiring looms. Spot welds. Machine changeovers.

Their app scores each trainee. How close was the tool to the mark. How many tries did it take. The score goes in a report and the report goes to the client.

Now the client asks for the same course in passthrough, so trainees can see the real bench. The studio ports it in a fortnight. The scores come back different, and nobody in the room can say why. Was the class worse, or was the mode different?

That is the whole problem. The score is meant to measure the person. Some of it is measuring the headset mode.

What you would sell them

Sell a calibration step that runs before the lesson starts. It is a 60 second scene. The trainee pokes a few dots, the app learns how far short or long that person is reaching today, and it hands the number to the lesson.

On Monday morning the trainee puts the headset on and pokes eight dots before the welding bay loads. The lesson then scores them against their own reach, not against a lab average. The trainer sees one extra line in the report that says how much the mode shifted this person.

Sell a wind-down at the other end too. When the headset comes off, the app tells the trainee to wait before fine handwork, and it says why.

What to charge

Test it at 300 dollars a month per app title, with the source included. That is a hypothesis and you should say so to the first ten studios you show it to.

Compare it to what the studio already spends. One engineer, one day, is the honest comparison. You know what a day of your own engineering time costs. If a studio would spend three days building this themselves and then maintaining it across headset updates, 300 dollars a month is a small line.

There is a second thing they already pay for, which is the argument with a client about a score. That argument costs a meeting, and sometimes a contract. Price against the meeting.

How you would build the first version

Build one scene, for one engine, for one headset. Four target distances, like the study used, at 20, 25, 30 and 35 cm. Eight pokes at each. Log where the finger stopped against where the dot was.

Store two things. The average miss, and how the miss grows with distance. Those are the two numbers the study found, so those are the two numbers your scene should return.

Use the headset’s own hand tracking. Do not add a camera rig. Do not build a dashboard, a cloud account or a trainer portal. Give them a file and a number.

Do not claim medical or safety validity. You are selling a measurement, not a treatment. The study did not test surgeons, and neither did you.

The one-week test

1. Day one. Build the scene. Four distances, eight pokes each, one output file. Nothing else.
2. Day two. Run it on yourself and four other people, in both modes. Look at whether your numbers move in opposite directions at all. If your tracking cannot see a difference of a few millimetres, stop here and fix that first.
3. Day three. List 20 studios and training vendors who ship headset courses in both modes. Find the person who signs off the scoring.
4. Day four. Send each one a two minute video of your five people and one question. Has a client ever asked you why the scores changed after a port? Offer the scene for 300 dollars a month, and say you are testing that price.
5. Day five. Count paid orders. Two paying studios out of 20 is a yes. Zero, with several replies saying they would build it in an afternoon, is a no.

What would kill this

The effect may be too small to sell. A lasting shift of 0.35 cm is about three and a half millimetres. That matters for a needle or a solder joint. It does not matter for grabbing a door handle. If your buyers only teach coarse work, they will not pay, and they should not.

Your tracking may be noisier than the thing you measure. The lab used a marker on the fingertip tracked 250 times a second. You will use the headset. If your own noise is larger than a few millimetres, your scene reports weather, not reach. Day two of the test exists to catch this.

The study cannot say why the modes differ. Scene, targets, hand appearance and the camera path all changed at once. In virtual reality people saw a drawn hand. In passthrough they saw their real hand. So a correction built on the wrong cause may make scores worse. Measure the bias and report it. Do not silently rewrite the score.

Studios may build it themselves. It is a small scene. The defence is not the code. It is having run it across many headsets, many builds and many people, so their number means something next to other people’s numbers.

The research may not hold outside the lab. One lab. One headset. Thirty-nine adults aged 18 to 31, most with little headset experience, poking flat dots on a table. Nothing there tells you what happens after an eight hour shift, with older workers, or on a different headset. If you build this, you are betting the effect survives all of that.

MEASURE THE MODE BEFORE YOU SCORE THE PERSON. That is the sentence you are selling, and it is true whether or not the exact millimetres hold up.

Just Out Tech explains new research in plain language. This article was drafted with AI assistance and checked by a human against the original source.