Drone Technology
A team of University of Tennessee students and an Oak Ridge National Laboratory researcher have created a robot with a brain. The vehicle — affectionately named Neon — looks like a cross between a Roomba and Disney-Pixar’s “Wall-E.”
It is capable of navigating a space while avoiding obstacles by sensing objects and then changing directions to avoid bumping into them.
The University’s neuromorphic research group created the bot as a step toward new drone technology for the Air Force.
Neuromorphic is a term that means “brain” to computer scientists. Robots typically get their brainpower, so to speak, from computers to which they are connected. Neon does not.
“Most computers run code,” said professor and neuromorphic group advisor Mark Dean. “There is no code in Neon. It has neurons and synapses and operates based on what it’s been trained to do.”
The prototype must still be booted up with a computer, but once it’s turned on, it needs no outside programming to move.
Student researchers Parker Mitchell and Grant Bruer demonstrated recently on a small laptop at the University of Tennessee’s Min H. Kao building.
Once its lights blinked on, they set the machine on the concrete ground in a maze of cones and plastic bins, occasionally stepping out in front of it to test its response time.
Treadwheels turning and occasionally shedding screws, the careful little robot whirred around the room with its goggle-like eyes pirouetting back and forth like one of the parading flamingos on BBC’s Planet Earth II.
Trained to fear collision, Neon learns as it goes, becoming smarter and performing with more accuracy each time it approaches something in its way.
Survival of the fittest
To get Neon to this point, many generations of the brain had to go through high-stakes training on ORNL’s flagship supercomputer, Titan, that might put some military boot camps to shame.
Former UT student and ORNL researcher Katie Schuman was the instructor.
“We ran it through a simulation where the next step is to train the brain not only to avoid objects but to target certain objects and it is incentivized to go as many places as it can, without hitting anything,” Schuman said.
The incentive Schuman mentioned is survival. Hundreds of thousands, maybe even millions, of versions of Neon’s brain perished at her hands during the robot’s 24-hour training period on Titan’s 18,000 computers.
Of course, robots can’t feel pain, but if they could, this would be a very sad story.
“It’s like survival of the fittest,” Mitchell said. “The good brains get to keep going, the bad brains die.”
That way, the brains evolve over time.
Mitchell and Bruer gave the small brain a working body, with the contributions of about 20 other engineering students in the neuromorphic research group.
Watching Neon’s various iterations, Schuman said, they started to notice interesting things about the way it learned. For instance, earlier versions would spin several times when they reached corners to get out of it. Now Neon only turns as far as it needs to go to change direction.
The robot also always turns the same direction to get out of corners or move away from obstacles.
“I think that’s an efficiency feature,” she said, “since its goal is to cover as much ground as possible.
Applications
Neon’s potential applications don’t stop at unmanned aircraft. Dean said autonomous robots with brains as small as Neon’s are rare, making the brain applicable for environments where resources are scarce and weight must be restricted.
That could be aerial drones, but it could also apply to satellites and underwater vehicles.
Dean said the next step is to train the brain not only to avoid objects but to target certain objects.
“The primary goal is to be able to build things that adapt to their environments,” he said, “so that they can survive in different environments based on what they learn.”
Source: Knoxville News Sentinel, by Brittany Crocker
The East Tennessee Economic Development Agency markets and recruits business for the 15 counties in the greater Knoxville-Oak Ridge region of East Tennessee. Visit www.eteda.org
Published January 31, 2018