The use of drones against Russia in the Ukrainian “Operation Spiderweb” was game-changingly effective. Somehow, they smuggled in not only hundreds of drones, but the pilots to fly them — the pilots operated the FPV drones from nearby, and then were smuggled back out of Russia lickety-split.
For their part, Russia is also deploying a spiderweb of drones against Ukraine, that spool up to 20 kilometers of fiber optic cable to physically connect pilot and drone to avoid radio signal jamming. There are now forests in the area criss-crossed with cables from drone missions past.
These ingenious schemes are the result of requiring one human pilot per drone. This is a legal requirement (for now) in the USA, as well as a technical requirement because of the way drones are traditionally built and flown.
Except it’s not a requirement. If you have one or more drones that can fly themselves, then you don’t need hundreds of pilots for your hundreds of drones, and you don’t need spools of fiber optic cable connecting drones to their pilots. AND if your autonomous drones are less than 250 grams in weight, then you don’t need one licensed human pilot per drone in the USA – one operator can oversee a hundred drones that fly beyond visual line of sight (BVLOS) with the right software.
But how can one operator monitor one hundred drones working together, even if AI helps coordinate their activity? All those video feeds, all the messages going back and forth, all the telemetry logs. A human, even a team of humans, does not have the mental bandwidth to keep an eye on a hundred drones’ logs in real time, with megabytes of logs every second.
All systems are go, “Are you sure?”
Control is not convinced, but the computer
Has the evidence, no need to abort
The countdown starts
The only way to scale an AI-coordinated fleet of drones is to also automatically monitor their telemetry and summarize status. What can quickly summarize many megabytes of data? A custom, designed-from-scratch, trained-on-drone-telemetry-alone summarizer large language model (LLM) that can tell you how many drones are operational, how is the mission proceeding, and what snafu is fubar.
And we at Astral.us made such an LLM for our fleet management logs, because despite all our pretty drones, we are software company at heart.