EXP-005
failed
AI for Autonomous Aerial Systems
Researching perception, reasoning, planning and natural-language interaction for UAVs operating in the physical world.
Survey in progress. No detections yet.
Problem
Most drones still need a pilot. I want to know what it takes to stop needing one.
Someone standing at the edge of a flood does not want a control stick. They want to say what they need and get an answer back. That sentence gets said once, under pressure, by a person with no technical vocabulary, and somehow it has to turn into perception, reasoning, planning, navigation and action on an aircraft with a finite battery and no signal.
"Fly over the flooded region, identify stranded people, map inaccessible roads, and report the highest-priority rescue locations."
Everything on this site works backwards from that sentence. What has to be understood, seen, decided and flown before you can answer it honestly. The console above is that sentence, executed: it surveys, it revises what it believes as the water rises, and it hands back a ranked list.
Can a spoken instruction become a flight mission that is safe and checkable without a pilot?
What does a UAV need to see, at what altitude and in what conditions, before its report can be trusted?
How should a UAV decide where to look next when its map is wrong and its battery is finite?
Can a firefighter or police officer use an autonomous aerial system on day one, with no drone training?
What is the largest model that can fly, and what does shrinking it cost on the cases you care about?
Can a UAV translate a high-level natural-language command into a safe, executable autonomous mission?
What perception stack gives a responder a picture they would actually act on, running entirely onboard?
How do several UAVs split an unknown search region when they cannot reliably talk to each other?