VR tracking answers one question many times every second: where is this object, and which way is it facing? The headset asks it about your head, the controllers ask it about your hands, and the answer arrives as a stream of positions that the software uses to render a scene from the right viewpoint. If that stream is accurate, fast and uninterrupted, VR works. If it is not, guests feel it immediately - as swimming, as lag, as a controller that jumps.
The important thing for a venue is that tracking is not a feature you switch on. It is a volume: a region of the floor where the system can see what it needs to see. Everything else in a machine specification - how many players, what game modes, where the machine can be placed - follows from the shape and size of that volume. Our companion guide on how virtual reality works walks the full sense-render-display-correct loop; this article stays on stage one, because stage one is where most paid sessions are won or lost.
There are two families of hardware that build the volume, and they fail in opposite ways:
Both are used commercially. The rest of this article explains what each one actually measures, where each one breaks on a busy floor, and what to ask before you sign a purchase order.
Tracking hardware does not report "the guest is facing the screen". It reports a pose: a position in three axes plus an orientation in three axes. That is six numbers, which is why the term is 6DoF (six degrees of freedom) - three for where, three for which way. Machines that only report orientation, so the guest can look around from a fixed point, are 3DoF, and the difference between the two is the difference between a cinema and a game. The full set of axes is broken down in our guide to what 6DOF means.
No single sensor produces that pose. A headset fuses two very different inputs:
Software then fuses the two into one pose stream. The practical consequence is that tracking quality is not just sensor quality: it is the quality of the sensor fusion, and it degrades gracefully when one input is missing. A headset that loses its optical reference for a moment keeps rendering from the inertial estimate, which is why a tracking fault often looks like a slow slide rather than a freeze.
Two numbers matter to a buyer. The first is the sample rate - how often a new pose is estimated. The second is the time between taking a measurement and showing the matching picture, which is the latency budget discussed later in this article.
Outside-in tracking is the system most people picture when they think of room-scale VR. Two small base stations are mounted in opposite corners of the play area, aimed inward. Each unit spins an internal rotor that sweeps a pair of infrared beams across the room; a headset or controller covered in photosensors records exactly when each beam crossed each sensor, and the timing gives its position. Published teardown figures put the rotor at roughly 3,600 RPM, which produces about 60 sweeps per second per axis.
The accuracy that comes out of that is the reason this family still appears in commercial products. Independent measurements reported jitter below 0.3 mm and positional RMS around 1.5 mm - sub-millimetre stability in a room-scale space, with no cameras on the guest. Valve specifies the practical SteamVR play volume as roughly 5 x 5 m with two base stations and about 10 x 10 m with four.
Outside-in wins where accuracy and repeatability matter:
The costs are equally clear, and they are all installation costs. The stations must be mounted high enough and angled to cover the play area, cabled or powered, and calibrated; a cable that was fine on day one becomes a fault call after a year of shopping-centre footfall. Line of sight matters: a base station cannot see through a pillar, and a person standing between a station and a controller blocks that measurement exactly as a body blocks a camera.
Inside-out tracking inverts the geometry: instead of the room watching the guest, the headset watches the room. Four or more cameras on the front and sides of the headset feed a SLAM (simultaneous localisation and mapping) pipeline that builds a map of visible features and locates itself inside that map, again fused with inertial data.
For a venue the appeal is deployment:
The costs are specific and worth writing down, because they are exactly the faults a venue sees in week three:
None of that makes inside-out the wrong choice. It makes the room part of the specification. A mat, a defined boundary and a few visual features are cheap, and they convert a fragile tracking volume into a reliable one.
In a demo room, occlusion is an inconvenience. On a paid floor it is a queue problem, because a lost pose usually ends the session rather than pausing it. There are four occlusion patterns worth designing against before installation day.
The design answer is not more hardware, it is a boundary. Define the play area with a physical edge the guest can feel and a virtual boundary the headset can show, keep the reflective surfaces out of the volume or matte them, and choose game modes that keep hands and guests inside the space they were given. That is the same design logic the wider station build uses, described in our guide to what a VR station is made of.
Tracking can be accurate and still feel wrong, because the picture must arrive while the movement is still happening. The measurement that matters is motion-to-photon latency: the time from the guest moving to the corresponding pixels leaving the display.
Peer-reviewed measurement work puts real headset latency in the range of roughly 21 to 42 ms, falling to about 2 to 13 ms for a well-learned task where the user has adapted to the delay. Manufacturer figures for one widely deployed commercial family are around 22 ms for the headset and about 31 ms for a controller. That gap between head and hand is worth remembering, because a guest perceives a lagging controller long before they perceive a lagging view.
The budget breaks down into stages you can discuss with a supplier:
| Stage | What happens | What it costs you |
|---|---|---|
| Sensing | Inertial and optical samples are captured. | Hundreds of poses per second; the guest cannot feel this stage alone. |
| Fusion and prediction | Sensor data becomes a predicted pose for the frame currently being drawn. | Prediction buys time back, but a wrong prediction appears as overshoot when the guest stops moving. |
| Render and transfer | Two eye views are drawn and sent to the headset. | The heaviest stage on PC-tethered systems; the reason a heavy scene is specified with the right computer, not just the right headset. |
| Display and persistence | The image is held on the panel for the exposure time. | This is where reprojection helps: it can re-warp an existing frame to a newer pose, hiding display lag. |
The crucial limit is what reprojection cannot do. It can correct the angle of a frame you already rendered; it cannot invent a pose you never measured. So when tracking drops out, no amount of clever display work will fix the experience - the guest sees the world move without their head moving, which is precisely the sensory conflict that makes people feel unwell. That is also why tracking faults and comfort problems are so often the same complaint.
Most tracking faults we see come from four avoidable setup mistakes rather than from hardware failure.
One more source of drift has nothing to do with the machine: the room changes. A new LED screen in the frontage, a seasonal decoration, a rearranged queue barrier. Any of those can move the reflective or occluding geometry around the play area, so tracking should be re-verified after every fit-out change rather than assumed.
You can evaluate tracking in about fifteen minutes on a demonstration floor, with no instrumentation. These five actions expose the failure modes above.
Write the results down next to the machine model. When you compare quotes later, this one page will separate a specification that was measured from one that was copied from a marketing page.
The table below turns everything above into the questions to ask, and the cost of accepting a weak answer.
| If the machine is | Ask specifically | What a weak answer costs you |
|---|---|---|
| Outside-in with base stations | Where do the stations mount, what is covered at guest height, and who re-aims them after a fit-out? | Corner dead zones and a monthly fault call; guests blame the game, not the geometry. |
| Inside-out standing platform | What is the controller field of view, and what happens when a hand leaves it? | Mid-session dropouts in the fastest moments of play, which is exactly when they are noticed. |
| Free-roam arena, either family | How many players share one tracked volume, and what is the occlusion plan when they cross? | Session restarts and a throughput ceiling lower than the one you were sold. |
| Any of the above | What is the measured motion-to-photon latency, head and controllers separately? | A comfort complaint you cannot diagnose, because you have no baseline number. |
| Any of the above | Which room surfaces in my building will interfere, and what do you recommend doing about them? | A tracking fault caused by your own floor, blamed on the machine. |
Read that list again as a purchase order rather than a questionnaire. A supplier who answers all five has already done the site engineering that makes tracking reliable; a supplier who answers none has sold you a headset.
For most venues the tracking decision and the machine decision are the same decision, because the machine is built around a volume of a particular size. Three placements cover the practical range.
Ask for the volume in metres for every option, not just the footprint of the cabinet, and check it against the guest mix you expect: adults need more clearance than children, and a group of four needs a boundary that keeps four people out of each other line of sight. Looking for a tracked space for a specific floor? The models below are specified with their tracking family and play area, and we will size the volume against your plan with a free 3D layout design.
The questions venue buyers ask about tracking, in the order they ask them.
The headset and controllers report a pose - position plus orientation in three axes each - many times per second. Inertial sensors detect fast rotation, optical tracking supplies an absolute position so errors do not drift, and software fuses the two into the pose stream the renderer uses. That stream has to be accurate, fast and uninterrupted for the experience to feel real.
Outside-in places base stations in the room that sweep infrared beams and watch sensors on the headset and controllers; it is very accurate and works in bright rooms but needs installation and line of sight. Inside-out puts cameras on the headset that map the room themselves; it needs no installation and scales with open floor, but controllers that leave the camera view lose tracking and featureless or very bright spaces are harder to read.
For the outside-in SteamVR family the manufacturer specifies about 5 x 5 m with two base stations and about 10 x 10 m with four. Inside-out systems are limited by the open floor you can give them and by how many guests share it, not by room hardware. In every case the usable volume is smaller than the room, because edges, obstacles and reflections cut into it.
Four causes dominate: occlusion (a body or object between the sensor and what it is watching), reflective or mirrored surfaces that create false readings, featureless and very bright conditions for camera-based systems, and setup errors such as base stations at the wrong height or angle. Consumables - low batteries, worn cables - surface as tracking symptoms too.
Measured motion-to-photon latency runs from roughly 21 to 42 ms in real headsets, with the controller typically slower than the headset. Reprojection can re-warp a rendered frame, but it cannot replace a pose that was never measured, so a tracking dropout shows the guest a world that moves without them. That mismatch is the same sensory conflict that makes people feel unwell.
Choose by deployment reality first. If the room is fixed, well-lit and the play area is defined, outside-in rewards you with accuracy and robustness. If you need to deploy quickly, move machines between sites, or open a large roaming floor without infrastructure, inside-out is the practical choice and the room - mat, boundary, surface treatment - becomes part of the specification.
Six lines to take into your next supplier conversation.
The next step is arithmetic on your own building: ceiling height, surface finishes, guest mix and the floor area you can give to a tracked volume. Send the plan through the free 3D layout design and we will mark out the usable volume and the machines that fit it, or start from the VR park solution overview for the full zone picture.
Tracking is a floor design problem. Review the VR park solution and send us your dimensions for a free 3D layout showing the tracked volume, the station shapes that fit it, and what each option includes.
