What Is IPD? How to Measure Yours, and Why Optics Change the Tolerance

IPD is the distance between your pupils. Here is how to find yours in thirty seconds, what a published IPD range does and does not tell you, and, for anyone who wants it, the optics underneath.

The short version.

• Your IPD is the distance between the centres of your pupils. About nine adults in ten fall between 57 and 70 mm.
• A published IPD range is not a list of the faces a product fits. On URXR One, 58 to 68 mm is the distance between the two optical axes, and how far apart they physically travel.
• So the arithmetic is simple. If your IPD is 72 mm and the mechanism stops at 68, your eyes are not 4 mm off centre. They are 2 mm off, one on each side. Every millimetre beyond the stop costs half a millimetre per eye.
• Whether 1 mm matters depends on the eyebox: how much room your pupil has to move before the image suffers. That is set by the optics, which is why two products with identical published ranges behave differently on the same face.

If that is all you needed, the next two sections tell you your own number and what it means. The physics is further down, clearly marked, for anyone who wants to check the reasoning.

Find your number

Ranked from most to least reliable.

Ask your optician. They measure with a pupillometer, and if you wear prescription glasses your PD is often already on file. Free, thirty seconds. Ask for the distance PD, not the near one. If your prescription lists two numbers, those are your monocular PDs, so add them.

Use a phone app built on the depth camera. Usually accurate to within a millimetre or two. Apps that use the ordinary camera and a credit card held against your forehead are considerably less accurate.

Measure it yourself, then correct it. Stand about 20 cm from a mirror, hold a millimetre ruler against your brow, close your right eye and line the zero mark up with your left pupil. Without moving the ruler, switch eyes and read where the right pupil lands. Three times, averaged.

Then add 3 to 4 mm. Your reflection sits about 40 cm away, so your eyes converge on it, and what you just measured is your near IPD, the smaller of the two numbers, and not the one that matters here. This is the step most guides leave out, and it is why self-measured figures run low. If someone can help, better: have them hold the ruler while you look at something at least three metres away.

Chasing accuracy beyond about 1 mm is not worth it. The adjustment mechanism on most hardware is coarser than that anyway.

Put your number in

Fig 1

Two separate things decide whether your eyes line up: how far the mechanism can travel, and how much room your pupil has once it stops.

Your IPD

63.0 mm

Drag to your own number.

How far apart the two optical axes can be set

Not adjustable — axes fixed at 63 mm 58 – 68 mm 55 – 72 mm

This is the quantity the URXR One range describes. If a product's range refers to a software shift instead, its axes do not move, so read the first option.

Published eyebox Region a whole 4 mm pupil must fit inside Your pupil

So will it fit me?

Three cases, and none of them is “check the range and stop.”

Inside the range. Set it and forget it. This is most people, most of the time.

A few millimetres outside. Halve the difference, and that is your real offset per eye. Three millimetres beyond the stop is 1.5 mm per eye, which most optics absorb without you noticing. What decides it is the next question.

Well outside. Now the architecture matters, and there is one question worth asking a vendor, a reviewer, or a friend who owns the thing:

When you are off centre, does the image clip, or does it soften?

Clipping means you lose the edge of the screen and it goes suddenly: your pupil position and your field of view are tightly coupled. Softening means a little sharpness and brightness at the periphery, gradually. A hard edge two millimetres away is a problem. A gentle slope two millimetres away is usually nothing.

That is a question you can answer on your own face in thirty seconds, and unlike a published number it means the same thing at every vendor.

Three more worth putting to any vendor before you buy:

• What does your IPD number measure? Ours is the optical axis separation, which is what makes the halving above work. It is a fair question to ask anyone, and the answer changes what their range means.
• Is IPD handled mechanically, in software, or not at all? Mechanical adjustment moves the optical axes, so your offset is measured from wherever the mechanism stopped. Software adjustment shifts the rendered image instead and leaves the axes where they are. That is real value, it fixes the stereo geometry and can meaningfully reduce eye strain, but it does not recover eyebox.
• Does the eye relief survive glasses underneath? Your own spectacles push your eye back, and the eyebox shrinks fastest exactly where the field of view is widest. A prescription insert that mounts inside the optics does not cost you this. Your own glasses do.

One scoping note: this piece is about display glasses, the lightweight tethered category you wear like eyewear. Full-size VR headsets sit in a different optical regime, and mixing the two is how most IPD explainers go wrong.

The physics, if you want it

Everything above is the practical answer. What follows is the reasoning underneath it: why the eyebox is set where it is, why architecture changes the answer, and where our own claims stop. Each section opens with its one-line version, so you can skim the bold and still get the argument.

What IPD actually varies by

In one line: the average is about 63 mm, the spread is about 4 mm, and two details usually get skipped.

Across a large adult sample the mean sits close to 63 mm with a standard deviation of roughly 4 mm.¹ That puts about nine adults in ten between 57 and 70 mm, with the full range running from around 51 to around 77 mm at the extremes. Women average a couple of millimetres lower than men. Children and younger teens are often below 56 mm, which matters if you are buying for a family.

Near IPD is smaller than far IPD. Looking at something close makes your eyes converge, so your reading IPD runs 3 to 4 mm narrower than your distance IPD. Display glasses render at a fixed virtual distance, several metres out, so distance IPD is the number that matters, and it is why the mirror method above needs its correction.

Faces are not symmetric. Your left and right pupils are often not equidistant from your nose bridge; opticians call these monocular PDs, for example 32 mm left and 30.5 mm right. Most consumer devices accept only a single combined number, so a strongly asymmetric face is always a small compromise. Worth knowing, so you do not assume something is broken when it is anatomy.

Two failure modes, and only one is fixable

In one line: bad geometry is a software problem; a pupil outside the eyebox is not.

Geometric mismatch. The device renders the two eye images for an IPD that is not yours. Stereo depth cues go off, the world can feel subtly wrong in scale, and your eyes work harder to fuse. Over a long session that becomes eye strain. Software fixes this.

Optical misalignment. Your pupil sits outside the eyebox, the volume where the optics deliver a full, clean image. This is hardware, and software cannot undo it.

Worth saying plainly: IPD is not the only thing that tires your eyes in a headset. Every device in this category renders at a fixed focal distance while your eyes verge at varying depths, and that conflict does its own work no matter how well the IPD is set. IPD is the part you can do something about, which is not the same as it being the whole story.

Why a stated range cannot tell you how it will feel

In one line: the eyebox is the number that would predict comfort, and there is no agreed way to measure it.

Eyebox is usually published as two numbers, 11 × 7 mm, horizontal by vertical.

Then the arithmetic that spec sheets never show. IPD error splits across two eyes: a device set to 63 mm against a 70 mm IPD leaves each pupil 3.5 mm off axis, not 7. On an 11 mm eyebox that is near the edge. On a 6 mm eyebox it is outside.

Three complications, none of which appear on a spec sheet:

• Nobody says which convention they used. Some vendors quote the range over which the centre of your pupil can sit; others quote a region your whole pupil must fit inside, which with a 4 mm pupil is smaller. An 11 mm eyebox therefore means either ±5.5 mm or ±3.5 mm of travel depending on whose sheet you are reading.
• Your pupil is not a fixed size. It runs from about 2 mm in bright conditions to 6 mm or more in the dark, so the slack you have watching a film in a dim room is less than the slack you have at a desk.
• Vertical is the one nobody discusses and the one that causes more complaints. A 7 mm vertical eyebox leaves very little room once a pupil is inside it, and how a frame sits on a nose varies by more than that between people. When fit fails in this category, it usually fails vertically first.

Why birdbath ties field of view and eyebox together

In one line: a 45-degree element inside a spectacle frame caps the aperture, and field of view and eyebox both have to come out of it.

Light leaves a micro-OLED panel above the eye, hits a beam splitter at 45 degrees, reflects off a curved semi-transparent combiner that doubles as the see-through lens, and passes back through the beam splitter to your eye.

Fig 2 · Birdbath

The outbound and return legs share one corridor, and a 45° element sets how tall that corridor can be.

1. 1Light leaves the panel and travels down to the beam splitter, which reflects it toward the world side.
2. 2The curved combiner, which also has to stay transparent because you look through it, reflects it back and does the optical work.
3. 3It passes back through the beam splitter to your eye, down the same corridor it came out of.

Two separate constraints operate here, and they usually get collapsed into one.

Constraint 1, the Lagrange invariant. The product of eyebox size and field of view is fixed by the panel size and the numerical aperture of the optics collecting from it. Widen the field of view and the eyebox shrinks, unless you bring a bigger panel or a faster system.

Constraint 2, plain geometry, and in eyewear it is usually the one that bites first:

eyebox ≈ clear aperture − 2 × eye relief × tan(half the horizontal field of view)

Eye relief is the term that does the damage, because it multiplies the field of view. Every extra millimetre of eye relief costs 2 × tan(half the horizontal field of view) millimetres of eyebox width, around 1.5 mm at wide-field-pancake angles, closer to 0.7 mm in a birdbath at half the field angle. A wider field of view does not merely shrink the eyebox. It makes the eyebox more sensitive to how far your eye sits from the optics, which is why wearing your own spectacles underneath hurts more than people expect.

Now put a 45-degree flat into that equation. A tilted plate presents a clear aperture of roughly its length divided by the square root of two, so buying aperture in a birdbath means buying depth and height in the front housing at once. Reaching the clear aperture a wide-field pancake module runs would need a housing several centimetres deep. That is not a design choice anyone declined to make, it does not fit on a face.

So birdbath designers have spent the budget on field of view and brightness, and the eyebox is what got smaller. Eyebox figures that have surfaced for this category, in optical papers, teardowns and occasional developer documents rather than on consumer spec sheets, run from roughly 14 × 8 mm at the generous end down to around 6 × 6 mm at the tight end, and the tighter figures tend to belong to the newer, wider, brighter designs. That is not a regression. It is the tradeoff being made deliberately.

The failure mode also differs from what people expect. Off-centre in a birdbath does not primarily blur the image. It clips it: you lose the corners of the virtual screen, one edge dims, and ghosting from the beam splitter becomes visible. Users describe it as “I can’t see the whole screen,” not “it’s out of focus.”

Two things to be fair about, since a comparison is coming. Pupil position and effective field of view are coupled in every non-pupil-forming headset, pancake included; architecture decides how much room you have before you feel it, not whether the coupling exists. And birdbath is not a bright architecture either: light crosses the beam splitter twice and reflects off a combiner that has to stay see-through, so a simple 50/50 birdbath delivers roughly a tenth of the panel’s output to your eye. Newer designs recover part of that with a polarising beam splitter and a quarter-wave plate.

Why mini pancake has more headroom

In one line: folding along the axis leaves the aperture nearly the full lens diameter, so the same budget starts from a larger number.

A mini pancake folds the optical path along the axis instead of around a corner, and it does it with polarisation. Light from the panel is polarised and made circular by a quarter-wave plate. It transmits through a half mirror on a curved surface, crosses the cavity, passes a second quarter-wave plate, and meets a reflective polariser in the state that element reflects, so it turns around. The half mirror reflects it forward again, and by now the accumulated retardance has rotated the polarisation into the state the reflective polariser transmits, so the light finally exits to your eye.

Three passes, inside a module 17 to 21 mm thick. The quarter-wave plates are what make the fold work at all: without them the reflective polariser would either always reflect or always transmit, and nothing would come out.

Fig 3 · Mini pancake

Folding along the axis, with polarisation doing the switching, and nothing diagonal in the way.

1. 1Polarised and made circular by the first quarter-wave plate, the light transmits through the half mirror, half of it lost here.
2. 2It crosses the second quarter-wave plate and meets the reflective polariser in the state that element reflects, so it turns around.
3. 3The half mirror reflects it forward again, another half lost. By now the accumulated retardance has rotated the polarisation into the state the reflective polariser transmits, so the light finally exits to your eye.

Three consequences for IPD tolerance:

• The aperture is not shared. No see-through requirement, no diagonal element eating the corridor. The usable aperture is close to the full lens diameter, so both constraints above start from a larger number.
• Degradation is closer to symmetric, and it is gradual. The stack is close to rotationally symmetric about its axis, so moving off centre costs sharpness and brightness progressively rather than clipping one edge. A gentle slope is easier to tolerate than a hard cutoff, even at the same nominal eyebox.
• Folding buys focal length per millimetre of thickness. Three passes let a module a couple of centimetres deep behave like a much longer optical path. It does not create optical power, that still comes from the curved surfaces, but it delivers a short effective focal length without a long barrel.

The practical result is that a mini pancake device can cover a wider slice of the population’s IPD from a single mechanical configuration than a birdbath of comparable field of view can. It is the architecture the URXR One uses, for exactly this reason.

What that does not mean

In one line: five places the headline oversimplifies, including two that count against us.

Pancake is not universally more forgiving. Against Fresnel lenses in a full-size VR headset, pancake optics are often the less tolerant choice, especially vertically. The advantage here is specific: mini pancake versus birdbath, at display-glasses scale, at comparable field of view. Drop that last qualifier and the claim stops being true.

Pancake costs you light, twice. Half is lost at the input polariser, because a micro-OLED panel emits unpolarised light. The fold itself then has a hard ceiling of 25 per cent, one half at the transmission and one half at the reflection, and real components land below it. Call it something under a tenth of the panel’s raw output reaching your eye. That is affordable only because micro-OLED panels can be driven very bright, and it is part of why these systems run warm.

Pancake has its own off-axis artefact. A reflective polariser has finite extinction, and quarter-wave retardance varies with both angle and wavelength. Light that leaks through on the wrong pass forms a defocused ghost, and because the leakage is angle-dependent, that ghost gets worse as your pupil moves off axis, not better. Pancake trades a hard clipping edge for a soft one, not for a clean one. Off-centre also shifts the distortion profile slightly, which some people perceive as the image swimming as they move.

Pancake is not see-through. The stack is opaque, so you are looking at cameras, not at the room. Whether that is a benefit or a cost depends entirely on what you want the glasses for. Different category, not strictly better.

More tolerance is not infinite tolerance. If your IPD sits at the far edge of the human distribution, check the supported range before buying anything, whatever the architecture.

Quick comparison

Restricted to the three architectures that actually ship in display glasses today.

Architecture	IPD tolerance	What off-centre looks like
Birdbath	Low, and drops as field of view rises	Screen edges clip, one side dims, ghosting from the beam splitter becomes visible
Mini pancake	Higher at comparable field of view	Gradual softening and dimming toward the periphery; polarisation ghosting rises off axis
Diffractive waveguide	High laterally, uneven	Image rarely vanishes, but brightness and colour shift across the field

The waveguide row looks like a counterexample to everything above, so it needs a footnote. Waveguides get a large eyebox by replicating the exit pupil across the grating, filling a bigger box with the same light. That is not a free lunch: overall throughput is very low, which is why these systems need extremely bright light engines, and the replication itself produces the brightness and colour non-uniformity. Their field of view is separately capped by the angular bandwidth the substrate can carry under total internal reflection. Waveguides are not beating the tradeoff. They are sitting at the other end of it.

What the published numbers actually look like

In one line: five products, five ranges measured against different things, and not one published eyebox, including ours.

Product	Optics	Stated IPD range	How it is handled	Frame sizes
Display glasses A	Birdbath	59.5 to 70.5 mm (11 mm span)	Software, 13 steps of 1 mm	One
Display glasses B	Birdbath	57 to 66 mm (M) / 66 to 75 mm (L)	Software steps	Two (pick M or L at purchase)
Display glasses C	Birdbath	57.5 to 69.5 mm (12 mm span)	No user adjustment published	One
Display glasses D	Birdbath	55 to 72 mm (17 mm span)	Automatic, but only with the separate companion unit	One
URXR One	Mini pancake	58 to 68 mm (10 mm span, 63 mm centre)	Manual adjustment	One

Competitor names withheld deliberately; all figures come from the manufacturers’ own published specifications and are easy to verify. Current at the time of writing.

The spans run from ten millimetres to seventeen, and these numbers are not comparable to each other. Read the fourth column and the reason is obvious. The widest range belongs to a product that cannot perform the adjustment without a separate companion unit. The narrowest belongs to a frame that is one of two you choose between at checkout. One vendor publishes no adjustment method at all.

Some of these ranges describe a mechanism. Some describe a comfort claim. They are measured against different things and then printed in the same units, exactly the kind of spec that looks rigorous and informs nothing.

The row needing two frame sizes is worth sitting with. That is the aperture constraint from Fig 2 surfacing on a pricing page. When one frame cannot stretch far enough, the honest engineering answer is to build two, and the customer absorbs it as a sizing decision made before they have ever worn the product.

Notice what no row contains. Not one product here publishes an eyebox, ours included.

It is worth being precise about why, because the obvious reading, that everyone is hiding something, is not quite it. There is no agreed way to measure an eyebox. One vendor can quote the region over which the full field of view stays visible; another where brightness has fallen by half; another a resolution threshold, each choosing the threshold. Some figures assume a nominal pupil, some do not. Some fix an eye relief, most do not state one. Published without its method, an eyebox cannot be compared to the eyebox printed beside it, and a category that started printing them would reward whoever measured most loosely.

So we are not going to print one and call it a differentiator.

Our own range is 58 to 68 mm, centred on 63, in a single frame. That is a statement about the mechanism: how far apart the two optical axes travel. On the table above it is unremarkable, and we are not going to dress it up.

The claim worth making is about the tolerance around it. Outside the range you are not shut out, you are offset, by half a millimetre per eye for every millimetre beyond the stop. Whether that offset is noticeable comes down to how much room your pupil has, and what happens when it runs out. On the first, a mini pancake starts from a larger aperture than a birdbath of comparable field of view can, for the reasons in Fig 2 and Fig 3. On the second: a birdbath clips, a pancake softens. A hard edge two millimetres away is a problem. A gentle slope two millimetres away is usually nothing.

That is something you can test on your own face in thirty seconds, which is more than a published eyebox would have given you.

Mechanical versus software adjustment

In one line: three approaches, and only the mechanical ones address both failure modes.

Continuous mechanical. A wheel or slider physically moves the optics. Best case, because it addresses both failure modes at once.

Stepped mechanical. Two or three fixed positions. Better than nothing; if you fall between steps, pick the nearest and expect a small compromise.

Software only. The rendered image shifts within the panel to match your IPD. This corrects the stereo geometry and can meaningfully reduce eye strain, which is real value. But your pupil is still off the optical axis, so it does not recover eyebox.

A device with a small eyebox and software-only IPD handling has fixed the cheaper of the two problems. Some devices with eye tracking measure your IPD automatically and either drive motors or apply the software correction, convenient, though worth confirming which one is happening.

None of this is exotic. IPD is a spec that gets published without context and then blamed for discomfort that had a fixable cause. Spend thirty seconds learning your number, understand that a published range describes a mechanism rather than a guest list, and every headset decision after that gets easier. For the two other specs that decide whether display glasses suit real work, see what is PPD and VST vs OST. For the full URXR One numbers, see the specs page.

¹ Dodgson, N. A., Variation and extrema of human interpupillary distance, Proc. SPIE 5291 (2004).