Mission brief
Identity and area in — the agent starts with a clear target profile.
The problem
Manual search is slow where minutes decide outcomes
Missing-person operations still rely on staged air assets, limited terrain access, and humans scanning endless video. Coverage gaps and repeat passes burn the first day.
-
2–8h
Typical delay before a full air search is flying and coordinated.
-
Blind spots
Teams re-cover the same ground while other sectors stay dark.
-
Fatigue
Human eyes miss weak matches across hours of aerial footage.
The solution
Same idea as a scent dog — airborne and autonomous
Hand over the area and the person’s identity cues. The agent owns the hunt from there: it chooses the path, remembers what it scanned, filters with ML, confirms with AI, then brings a human in with live video and exact location.
Self-directed path
The AI decides where to fly next from live results and terrain.
Search memory
Covered sectors stay marked — no blind repeat passes.
Human when found
Operators get the stream, fix, and confidence — then teams move.
Mission flow
How the AI runs the search
Fully automatic from brief to find — humans join at verification and dispatch.
-
Brief the agent
Mission + person ID — the “scent”: clothing, features, last knowns.
-
AI plans the path
The model chooses the search pattern and adapts as evidence comes in.
-
Remember & advance
It logs what was scanned and skips ground already covered.
-
ML match → AI verify
Fast ML filters for match parameters; AI confirms the strong candidates.
-
Human in the loop
Alert with live stream and location — search teams get the handoff.
Detection pipeline
Fast ML first. AI confirmation second. Human last.
Frames are scored against the brief in real time. Only strong hits reach AI verification — then a human gets the stream, map fix, and handoff to the ground team.
Input
Brief
Mission + person ID
Stage 1
ML match
Fast parameter score
Stage 2
AI verify
Confirm the candidate
HITL
Human
Review & decide
Output
Teams
Stream + location
Weak ML scores stay in the search loop. Only verified finds interrupt a human.
System architecture
Cloud brain. Drones in the field. Teams on the ground.
The AI agent runs in the cloud: it plans, remembers, and identifies. Drones stream video and telemetry up; commands and search paths come back down. When a find is verified, operators and search teams get the live handoff.
Cloud
AI search brain
- Mission agent
- Search memory
- ML match
- AI verify
Field
Drone fleet
Autonomous flight · onboard capture · thermal ready
↑ video & telemetry · ↓ flight commands
HITL
Operator console
Brief mission · review finds · STOP / override
↔ live stream · confirm handoff
Ground
Search teams
Receive fix, confidence, and full live stream
← dispatch when AI + human agree
Cloud plans and remembers. Drones execute and sense. Humans decide. Teams recover.
Why it matters
Autonomous coverage. Human judgment on the find.
The agent burns search time so people do not. Operators stay focused on decisions, not joystick paths.
Minutes, not hours
Wheels-up fast; the model keeps searching without waiting on radio relays.
No wasted sky
Memory of covered sectors means every pass adds new ground.
Any terrain, day or night
Drones and thermal-ready vision reach places foot teams cannot.
Technology
Agentic flight. ML triage. Cloud vision.
Autonomous aircraft under an AI mission agent — computer vision, deep learning, and real-time streaming from edge to cloud so identification and handoff stay continuous.