Spatial Computing
Software that knows where things are.
Make space itself part of the interface. We design spatial applications that map environments, keep content in place and share it across users, connecting scene understanding to the 3D and operational data you already hold.
The opportunity
Start with a decision
worth improving.
A facilities team wants service notes, sensor readings and routes to stay exactly where they belong in a building, visible to whoever arrives next on any supported device.
From possibility to a working system
Follow the flow.
Three connected decisions. One considered architecture.
Conceptual flow, not a live system or measured result.
Map the space
Capture the environment as meshes, planes and recognized objects. Decide what must be scanned, what comes from existing CAD or BIM models and how often the map needs refreshing.
Persist and share
Place anchors that survive between sessions and can be resolved by other users and devices. Define who may create, see and remove spatial content, and where that data is stored.
Design natural input
Combine hand, eye and voice input with clear feedback and fallbacks. Treat gaze data and room scans as sensitive, with explicit retention rules.
The work, made concrete
What we can build
with your team.
We agree scope, dependencies and acceptance criteria before delivery. Your project can start with an assessment, a pilot or an integration into an existing system.
Spatial data and mapping strategy
Clarify the system boundary, the responsible owners and the decisions the architecture needs to support.
A persistent, shared-anchor application
Turn the agreed design into reviewable work, evaluated against representative inputs and explicit success criteria.
Interaction design and spatial data governance
Make the next step operable: document responsibilities, known limitations and the path from pilot to ongoing use.
In depth
Go deeper into spatial data
Two pieces of plumbing decide whether spatial content holds up in practice: the pipeline that brings building models onto devices, and the anchors that keep content in place for everyone.
- 01ProcessBIM to headset workflowA step-by-step BIM to AR workflow: export to IFC, optimize geometry, convert to glTF or USDZ with element IDs, align to survey control and govern versions.
- 02ArchitectureShared persistent spatial anchorsHow persistent, shared spatial anchors work on ARKit, visionOS, ARCore, Meta Quest and OpenXR, with a reference architecture, data model and privacy rules.
What a spatial application actually knows about a room
Spatial platforms expose a few kinds of information about the environment and the user. Knowing which ones an idea depends on is the quickest way to judge whether it is feasible on a given device.
- Scene mesh
- A triangle surface reconstructed from the device's sensors, used to make content collide with walls and furniture or hide behind them.
- Plane
- A detected flat surface such as a floor, wall, table or ceiling, with an outline and orientation, on which content can be placed.
- Scene classification
- Labels the platform attaches to surfaces or mesh regions, such as seat, door or window, so content can react to what a surface is.
- World anchor
- A fixed pose in the room that the system keeps tracking, used to hold content in place; persistence and sharing are covered on the anchors page linked below.
- Hand tracking
- A skeletal model of the user's hands, giving joint positions for pinches, grabs and pointing without a controller.
- Gaze targeting
- Using where the user looks to choose what an action applies to, usually confirmed with a pinch, a tap or a spoken word.
- Reference geometry
- Existing CAD or BIM models of the space or equipment, imported to give the application precise shapes the sensors cannot capture.
The spatial data stack beneath an application
- Interaction
Hand, eye and voice input, with feedback and fallbacks for each.
- Scene understanding
Meshes, planes and classifications produced live by the device.
- Anchors and frames
Anchors, site coordinates and the transforms that tie them together.
- Spatial content
3D assets, CAD and BIM geometry, annotations and their versions.
- Operational data
Asset, sensor, work order and building systems the content refers to.
Scan the space, import the model or combine both
Every spatial application needs a representation of the place it works in. Where that comes from shapes accuracy, effort and upkeep.
- If
The space is small, changes often and no model of it exists, such as a meeting room or a temporary work area.
ThenRely on live scene understanding from the device, refreshed every session.
Scanning costs nothing to maintain and always reflects the room as it is now.
- If
An accurate BIM or CAD model exists and the space matches it, such as a recently completed building.
ThenImport the model as reference geometry and align it to the site with surveyed points.
The model carries precise geometry and element data that live scans cannot provide, such as what sits behind a wall.
- If
A model exists but the space has drifted from it through refits and unrecorded changes.
ThenCombine both: use the model for structure and identity, and live scans to show where reality differs.
The differences are often the most valuable information, for example before a retrofit.
- If
The site is large or outdoors, such as a campus, yard or plant.
ThenStart from survey data or a geospatial reference and add device-level anchors for detail.
Device scans struggle to stay consistent across large areas without an external reference frame.
Choosing input methods for spatial interfaces
Good spatial interfaces combine inputs, so each one covers the others' weaknesses.
| Factor | Hands | Eyes (gaze) | Voice | Controller or touch |
|---|---|---|---|---|
| Precision | Good for grabbing and moving; weaker for fine selection | Fast for choosing between targets; poor for fine positioning | Exact for named commands; useless for pointing | Highest, with physical buttons or a screen |
| Fatigue in long sessions | Raised arms tire quickly; keep interactions low and brief | Low, if targets are large enough to look at comfortably | Low | Low |
| Gloves, noise and dirt | Thick gloves can defeat hand tracking | Unaffected by gloves; safety eyewear may interfere with tracking | Struggles near loud machinery | Works with gloves if buttons are large |
| Accessibility | Excludes some users with limited hand or arm movement | Helps users with limited mobility; needs alternatives for others | Helps users who cannot use their hands; not for speech impairments | Familiar; needs at least one free hand |
| Privacy sensitivity | Hand images and movement patterns | High: gaze can reveal attention and interest | Recordings of speech and bystanders | Low |
Device support differs; confirm which inputs your target platforms expose to applications, and in how much detail, before designing around them.
Treating scans, meshes and gaze as sensitive data
Spatial applications collect information people rarely think of as data. A scene mesh of an office reveals its layout and contents. A scan of a home shows how someone lives. Gaze data shows what a person paid attention to and for how long. Each can be personal, commercially sensitive or both, and each is easy to retain by accident because it arrives as part of normal operation.
Decide what the feature actually needs before collecting anything. An app that places furniture needs planes, not a stored mesh; an app that highlights the button a user looks at needs the target, not a gaze log. Keep raw sensor data on the device where the platform allows, store only derived results, and set retention periods for whatever reaches a server. ColdAI treats gaze data and room scans as sensitive in its spatial work, with explicit retention rules1.
Workplace use adds a further question: whether spatial data could be used to monitor employees. Write down what the data will and will not be used for, involve employee representatives where your jurisdiction expects it, and take advice before using attention or movement data to judge individual performance.
Frequently asked questions
What is the difference between spatial computing and augmented reality?
Augmented reality describes what the user sees: digital content laid over the real world. Spatial computing describes what the software does underneath: mapping the space, understanding surfaces and objects, keeping content anchored and taking input from hands, eyes and voice. Most AR applications rely on spatial computing, but spatial computing also powers mixed reality apps and shared spaces that go beyond a simple overlay.
Do spatial computing apps need a headset, or do phones count?
Phones and tablets count. Modern phones map surfaces, track images and hold anchors, which is enough for placement, wayfinding and many shared experiences. Headsets add hands-free use, hand and eye input and a wider view of content around the user. Start from the task: if people need their hands free or must work with content around them for long periods, a headset may justify its cost.
How often should a scanned map of a building be refreshed?
Whenever the space has changed enough that devices struggle to recognize it, rather than on a fixed calendar. Watch the signals: anchors that take longer to resolve, rising failure rates in particular areas and user reports of misplaced content. Spaces that change daily, such as warehouses, may need routine re-mapping of specific zones, while stable areas such as plant rooms can go much longer.
Can spatial content be linked to live IoT sensor readings?
Yes, and it is one of the more useful patterns. Each sensor or asset is given a location in the spatial model, and the application fetches its current reading when the user looks at or approaches it. The location mapping and the data feed should be managed separately, so a sensor can be moved or replaced without rebuilding the spatial content. Our IoT and smart cities practice covers the sensor side.
Can eye tracking show what workers look at during a task?
Technically it may, depending on what the platform exposes to applications, but it is one of the most sensitive uses of spatial data. Attention data can reveal fatigue, interest and health-related signals. If there is a legitimate purpose, such as evaluating a training scenario, use aggregated and anonymized results, tell participants exactly what is captured and take legal advice first.
Sources
Connect the architecture
The next connection.
Explore the complementary capabilities that can turn an individual technology into a complete workflow.
Your next move
Bring us the
real problem.
A workflow that takes too long. A system that cannot connect. An idea that needs a technical path. Start there, and we can define what to investigate, build and measure.
- A location-bound workflow or information need
- Existing CAD, BIM or site data where available
- A target device and platform roadmap