Live spatial reconstruction
Rebuilds the space as a 3D mesh while the user looks around, and streams a lightweight version back to the device.
A KAIST ARRC research project · Early access comingSpeak to your space.
Symbiotic Augmented Intelligence for Reality
Speak to your space.
Reality responds.
SymbAIR understands the space around you in real time and turns a single sentence into an XR experience placed right in it.
“
Symbiotic
People, AI and space move as one loop
In SymbAIR, people express intent, the device senses reality, and AI understands and plans over that space. The result comes back as an experience placed in the user's own surroundings. No part replaces another; each does what it does best, together.
SymbAIR
People
express intent
Space
is sensed
AI
understands & plans
World
becomes experience
How it works
From sensing to experience, in real time
01
Sense
Sense
Camera, depth and pose streams from the headset capture the space the user is in, as it is.
02
Understand
Understand
The space is reconstructed in 3D, and its objects and their relations become a scene graph — like “keyboard on desk”.
03
Plan
Plan
An AI agent interprets voice or text commands and decides what to do, and where, grounded in the understood space.
04
Realize
Realize
The decision reaches the device, is rendered with physics, and appears in the user's real surroundings.
Capabilities
Core capabilities for spatially aware XR
Scene graph understanding
Recognises and names the objects in a space, and keeps the relations between them as structured knowledge.
Natural-language commands
Interact with your space through Korean and English voice or text. No menus to hunt through.
Object-relative placement
Turns phrases like “to the right of the keyboard” into exact positions, relative to where the user is looking.
Many users, one space
Several headsets join the same space at once, with each user's state and results kept cleanly apart.
On-device + edge hybrid
Latency-critical work runs on the device; heavy perception and reconstruction run on an edge server.
Platform
Three layers, one platform
SymbAIR is designed so that device, edge and agent layers each focus on their own job — and new capabilities and devices extend it layer by layer.
Agentic Layer
Agentic layer
An AI agent understands language and composes the platform's capabilities into experiences in space.
Edge Intelligence
Edge intelligence
A GPU server handles the heavy work — reconstruction, object recognition and relation reasoning — in real time.
Device Runtime
Device runtime
Tracking, hand input, rendering and physics run natively on the device, which stays the source of truth for moving objects.
Real time first
Sensor streams stay off the AI's reasoning path, protecting the experience's responsiveness.
Extend by capability
A new capability becomes usable by the agent simply by registering it with the platform.
Per-user state
Space and user state are kept apart, so results never mix in multi-user sessions.
Use cases
What understanding space makes possible
01
Education & training
Recognise real equipment and spaces, and overlay step-by-step guidance and hands-on content on top.
02
Industrial sites
Reconstruct workspaces and structure their objects to tie inspection and maintenance data to real locations.
03
Spatial collaboration
Several people place and review virtual objects together, in the same physical space.
04
XR & AI research
An experimental base for research on spatial understanding, agents and human–AI interaction.
Roadmap
From research to product
Now
Spatially aware XR
- Live spatial reconstruction
- Scene-graph objects and relations
- KO/EN commands with object-relative placement
- Multiple headsets at once
Next
A shared World
- A World that many users see together
- Experience memory: record and retrieve spaces and experiences
- Replaying and adapting recorded experiences
Vision
Generative XR experiences
- Experiences generated from knowledge and asset retrieval
- AR glasses, mobile and other devices
- Platform access for partners
Collaborate
We're looking for partners
From joint research to pilots in the field — tell us what you're interested in and we'll get back to you.
Research partnership
Joint research, funded projects, data and experiments
Early access · pilots
Evaluation, field pilots, commercial partnership