Technology
A brain that flies with the drone
Perceive. Lock. Simulate. Decide. All on board.
- 01
Perceive
Spots landslides, cracks, floods and defects on board.
- 02
Lock
Grades the risk to inspection standards, with a confidence score.
- 03
Simulate
A physical world model plays out what happens next.
- 04
Decide
Picks the safest action: re-plan, alert, close.
World model
It imagines before it acts
Real physics runs as the brain’s imagination. Decisions rest on consequences, not just detections.
- Rigid-body dynamicsSpan failure, vehicle fall
- Finite elementsSkin crack tearing under pressure
- Flood flowLevee breach, inundation
- Fatigue · Paris lawLoose bolts, crack growth
Architecture
Decide where it is safest to decide
Routed by confidence. If the network fails, the drone still decides.
> 0.85Acts on its own. Flies with no signal.
0.60 – 0.85Nearby double-check. Data stays local.
< 0.60Hard cases and updates. Never critical.
Targets
First-generation engineering targets
<30 ms
On-board inference
Today: >500 ms
≥75%
Tasks done on board
Today: 30–50%
≤1 s
Flight recovery
Today: seconds
≥90%
Missions on weak links
Today: 83–90%
5–15 W
On-board power
0.45–2B VLA, INT4/8
3 min
200 m slope survey
On foot: 1–2 h (demo test)
To be verified with third parties.
Why we win
Compute is cheap. Understanding hazards is not.
Unclear → clear
Scene knowledge
One standard for every hazard.
Costly → affordable
Light model on board
0.45–2B VLA, quantised.
Late → in time
Offline closed loop
Stable flight in about a second.
One-off → reusable
Shared platform
Live digital twin, routine to emergency.