Sandia National Labs developing AI drones for national security
Researchers are testing artificial intelligence with drones, teaching them how to ‘play tag’
Sandia National Labs is testing artificial intelligence through drones in a project aimed at enhancing national security.
Lead researcher Spencer Jensen demonstrated the project, called AutonomyNM, by playing “drone tag.”
“The rules of drone tag are simple. The idea is that the blue drones try to get the red drones to crash into each other, and the red drones are trying to tag the blue drones,” Jensen said.
While the drones appear to be playing a game, the research has broader implications.
“We have critical infrastructure everywhere. So that’s obviously a national security issue for us,” Jensen said.
The work could eventually support defense systems, including power plants, nuclear waste sites, and other components within the nuclear sector.
Researchers at Sandia Labs are conducting numerous tests to teach the drones using positive reinforcement.
“Reinforcement learning is really just trying to learn from experience. So it’s a lot like what humans do and what animals do. It’s a lot like how you would train a puppy,” Jensen said.
He explained that the process helps develop human-like instincts in the drones.
“The nice thing about reinforcement learning is it’s not specific to drones. It can be applied to anything like a ground vehicle, it might be a little safer than a drone,” Jensen said.
The drones and vehicles are learning through artificial intelligence, though Jensen acknowledged there are still challenges to address.
“I think we’re getting there. I think we’re working out some of those bugs, and I think this research is part of that and to say, can we actually trust these algorithms and how can we validate them as well?” Jensen said.
Researchers said the next step is to move the drones out of the lab and test them in real-world settings.
Click Here For The Original Source.
