Overlaying the Model on Reality

Registration is the technical name for aligning digital content to the real world so it appears in the correct place, at the correct size and orientation, and stays there. It is the invisible foundation under every AR and MR experience: get it right and the model sits believably on the table or in the room; get it wrong and the design floats, tilts, or slides away from where it should be. There are a few main ways devices do it, and it helps a designer to know them at a working level.

Markers and targets. The oldest and most reliable method: the device recognises a known printed pattern - a coded square, or a specific image - placed in the real scene, and because it knows the pattern's real size, it can work out exactly where that pattern is relative to the camera and lock the digital model onto it. Put the marker on your table or on the base of a physical model and the digital content registers to it. Reliable, but it needs the marker in view. Surface and environment tracking (SLAM). Modern headsets and tablets use simultaneous localisation and mapping - as the device looks around, it identifies distinctive features in the real scene, builds a rough three-dimensional map of the surfaces (the floor, the walls, the table), and simultaneously works out its own position within that map. This is what lets a model rest on your real floor with no marker at all. Placed anchors. Often you simply position the model by hand onto a known real point and orientation - line up a corner of the model with a real corner of the room - and the device remembers that anchor and tries to hold the content there.

In practice a device blends these, continuously fusing camera images with motion sensors many times a second to keep its estimate of where it is and where the model belongs. When it works, the effect is magical: the design simply sits in the world. But notice what has quietly happened - the position of your design is now the output of a live estimation process running on noisy sensors and an imperfect map, not a measured, surveyed coordinate. That distinction is the whole of the next two sections, and the whole reason the overlay can never be the source of truth.

The reason AR alignment is never perfect is not that the hardware is cheap or immature; it is structural, and it will persist even as devices improve. The device is estimating its own position and the shape of the room from imperfect information - camera images that blur and have limited resolution, motion sensors that accumulate tiny errors, a feature map of a room that is approximate and can be fooled by blank walls, shiny surfaces, changing light or a scene that moves. Every one of those inputs carries noise, and the position of your model is computed from all of them, so the model inherits their error.

Two failure modes matter for design. Registration error is the static mismatch: at any instant the model may sit a little off its true place - a few millimetres in ideal close-up conditions, easily several centimetres or more at room and building scale, and worse the further you are from where alignment was established. Drift is the error changing over time and movement: as you walk across the room and back, the device's running estimate of where it is accumulates small mistakes, so the model that sat correctly on the near wall has crept off the far wall by the time you reach it, or has shifted when you look back. Tracking can also momentarily lose its lock - fast movement, a featureless surface, a dark or over-bright scene - and the content jumps or swims before it recovers.

These are not bugs awaiting a patch; they are the nature of estimating position from sensors in the real world, the same class of problem that makes any tracking system uncertain. Good hardware and software reduce the error, sometimes impressively, but never to zero and never reliably to survey grade across a whole site. For the design job this is completely acceptable: to judge whether a corridor feels tight or a layout works, a few centimetres of wander is invisible and harmless. For a build job it is disqualifying: a few centimetres is the difference between a wall in the right place and a wall that must be knocked down. The lesson is not that AR is broken - it works beautifully for what it is for - but that its position is an estimate with honest error and drift baked in, and a design tool must be used in full knowledge of that.

Put the two previous sections together and the conclusion is unavoidable, and it is the spine of this whole module: AR and MR overlay is a superb aid to understanding and cannot be a source of binding truth. The overlay's job is to help a human see and judge a design in place - and for that, an estimated position with a few centimetres of error and some drift is more than good enough, because human spatial judgement is itself coarse and contextual. You are asking does this feel right, does this clash, does this view work, and those answers do not change if the whole scene is a couple of centimetres out. The overlay serves that beautifully.

Binding results are a completely different kind of question, and they demand a completely different instrument. Where exactly is the boundary? At what level do we dig? Where precisely does this wall set out? Is this structural clearance adequate? These need certified, repeatable, legally accountable accuracy, and they are answered by instruments and people built for exactly that: the total station and the licensed surveyor, working from the verified drawings and measured survey data, under the National Building Code of India and local rules. A total station measures angles and distances to millimetre precision against fixed reference points and is trusted to set out buildings; an AR headset estimates position from noisy sensors and is trusted to make a space feel real. They are not competitors, and the headset is not a cheap substitute for the instrument - they answer different questions, and confusing them is how mistakes with real cost and liability happen.

So the discipline is simple to state and must be held firmly: use the overlay to understand, communicate and judge; take every binding, measured, set-out result from the survey. On site this means AR can helpfully show a worker roughly where the services will run so they grasp the coordination, while the actual marks they build to are set out by the surveyor. In the studio it means the tabletop or in-room model persuades and clarifies while the drawing set remains the record. This is not a grudging caveat; it is the correct, professional way to get the real value of the overlay - its power to make design legible - without ever borrowing an authority it does not have. The overlay shows you the idea in place. The survey tells you the truth.

Knowing the overlay is approximate does not mean distrusting it uselessly; it means using it with a few habits that keep it honest and get the most from it. Good AR and MR practice is largely about managing registration deliberately rather than pretending it is perfect.

Anchor to a known point and re-check. Align the model to a clear, real reference - a marked corner, a column gridline, a surveyed setting-out point - rather than dropping it vaguely into the room, and re-align if you have walked far or the content looks like it has crept. The closer you work to the anchor, the smaller the error. Give the tracking what it needs. SLAM works better in well-lit, textured, static scenes; blank walls, mirrors, glass, dust, harsh backlight and moving people all degrade it, all common on Indian sites, so expect more drift outdoors and in empty concrete shells and trust the overlay less there. Sanity-check against reality. Where you can, compare the overlay against something real and known - does the virtual floor sit on the real floor, does a virtual gridline land on a real column - and treat any visible mismatch as a reminder of the error budget, not something to measure from. Keep ground truth in the drawings. The model in the headset should always trace back to the verified BIM and drawings; the headset is a viewer of that truth in place, never the master copy.

Above all, never close the loop from overlay to build. The overlay can tell a designer or a client a great deal - this is roughly where it goes, this is how it will feel, this is the coordination - and that is genuinely valuable on site as well as in the studio. But the chain from a measured drawing to a set-out mark on the ground must pass through the surveyor and the total station, not through the headset, every time. Used this way - anchored deliberately, its limits respected, its ground truth in the drawings, and the binding line always deferred to the survey - overlay becomes exactly what it should be: a powerful, honest aid to seeing design in the real world, and never a shortcut around the instruments that hold the truth.