US Navy Developing State-of-The-Art Drone AI


This post is also available in: עברית (Hebrew)

The U.S. Navy is looking for a drone that can fly over a damaged airfield and come up with a plan to get planes back in the air quickly, the head of the Navy Expeditionary Combat Command (NECC), Rear Adm. Brian Brakke, told recently.

“How can I use UAS to run the initial damage assessment across that airfield?” he asked, outlining his thinking. “Is there artificial intelligence, deep machine learning, that can run across that field, scan it, tell me what munition I have left, what condition the airfield is in; then map out a plan that my teams can look at and say: ‘OK, this is how we’re going to attack this problem’?”

Experts say the technology for Brakke’s notional drone is partly available already — there are drones that can create this kind of maps of locations and send them back for analysis. But getting the drone to also come up with a plan of attack for repairing the damaged airfield is tricky. “We’ve got commercial drones flying right now and taking volume measurements of stockpiles,” said Michael Blades, an analyst with consulting and research firm Frost & Sullivan, who studies unmanned systems. “That kind of high-precision mapping and survey is going on right now. You can get a good three-dimensional map by flying over a few times.” The issue, Blades said, is the analysis piece — telling the difference between a rock and an unexploded ordnance is still very much reliant on expert human eyes. “To do that autonomously, that’s the leap,” he said.

Michael Horowitz, an unmanned systems expert at the University of Pennsylvania, agreed that what’s out there now isn’t exactly what Brakke is looking for. “Current technologies could give you a high-fidelity map after an attack,” he said. “The ability to use that data to generate a blueprint to repair that runway would be a new step.”

Developing the AI to the level it can help operators plan airfield recoveries could range from relatively simple to enormously difficult, Horowitz said. An easier hurdle would be creating an algorithm that had four pre-programmed damage scenarios and recovery plans, and based on what the drone is seeing it could then recommend the best course of action.

Whatever technologies get delivered to NECC, Brakke said companies need to factor in maintenance and support needs. “If I don’t have the folks to work on it, to understand it, to maintain it, I’m not creating a more effective force, I’m creating something that’s going to sit on a shelf that we’re not going to use,” he said.