Arkin Studies Animals to Build Smarter Robots

Jul 3, 2014 | Atlanta, GA

Related Media

Click on image(s) to view larger version(s)

  • Ronald Arkin

  • Wolf

For More Information

Josie Giles
IRIM Marketing Communications
josie@gatech.edu

This story was originally published in the Georgia Tech Alumni MagazineVol. 90, No. 2, 2014

Can studying the mating behavior of birds help the U.S. military develop better unmanned systems? That’s what Ronald Arkin, a roboticist at Georgia Tech’s College of Computing, and other researchers aim to find out as part of the U.S. Navy-funded Heterogeneous Unmanned Networked Teams (HUNT) Project.

Initiated in 2008, the HUNT Project is a multi-phased study that looks at assorted animal interactions—from wolves stalking an elk to squirrels hiding acorn caches—as inspiration for developing new algorithms to guide intelligent autonomous systems. For now, Arkin has been working with computer models and little bots in the lab. But things can always scale up to larger, more robust unmanned vehicles.

“That’s the beauty of the basic research,” he says. “It’s not limited to a physical type of platform.”

One of the earliest subjects of HUNT was “lekking” behavior in birds, in which a group of males gathers around—but not too closely—a very handsome specimen (a “hotshot”) in order to mate with females. This became the basis for seeing how one could distribute autonomous systems behind enemy lines “without using strict formation control” but in a way that “maximizes the likelihood of encounter” with the enemy, Arkin says.

In 2010 and 2011, Arkin and his team moved on to wolf packs. Initially, they thought the wolves coordinated with each other when hunting elk. But Dan MacNulty, a professor of wildlife ecology at Utah State University, disabused them of that notion. “When we brought Dan in the first time, he informed us that there is no coordination,” he says. “They are all individual, greedy agents.”

So how exactly did they work as a pack without explicit rules or communication? One possible explanation was that a predator chasing down an elk indicated to the others that the hunted animal was weak. So applying a probabilistic model to the stage of a hunt, Arkin tried to “replicate that behavior in robotic systems to see if we could do the same sort of thing both in simulations and platforms.” And he succeeded.

Following on the wolf pack research, Arkin then looked at bird mobbing, in which birds gather to drive off a stronger predator. Did it make sense for a weak bird to feign strength and participate in the mobbing? His simulations demonstrated that under certain conditions, yes, it did. And those same lessons could be applied to a low-power robot or one that’s out of ammo.

Arkin is now looking more broadly at robot deception. But, he explains, ultimately all of the pieces of HUNT relate to one another as examples of biologically inspired group behaviors.

Related Links