Team Teaches Drones How to Fly Like a Flock of Starlings

Primary Investigator: Umit Uyar

Anyone who has watched a flock of starlings or pigeons dance in the sky has to marvel at how so many individuals can fly in such precise patterns without colliding and, seemingly, without communicating with one another.


A team led by computer engineering professor Umit Uyar of The City College of the City University of New York is working to teach drones how to do it. Such drones – actually, they’re autonomous unpiloted aerial vehicles or UAVs – could work independently and yet together to clutter and confuse enemy radar or take out a hostile jamming station.


The U.S. military is keen to make use of such swarming technology, especially if it could use cheap or even disposable unpiloted aerial vehicles. With a grant from the U.S. Army Combat Capabilities Development Command, organized by the MSI STEM Research & Development Consortium, Uyar and colleagues have taken the first steps towards programming artificial intelligence into UAVs to that they  can act independently and yet collectively.


“Unlike drones, which are controlled remotely, these govern themselves,” Uyar said. “The pilot is the artificial intelligence software, which running inside the drone computer. There is no human intervention, no centralized control.”


This allows the UAVs to react quickly, usually within milliseconds. “They adapt as they go, without control or preplanning. If there are changes in the mission, they adapt to it without even being instructed to do so,” Uyar said.


Uyar and Janusz Kusyk, a professor and artificial intelligence and game theory specialist at the New York City College of Technology, are working together to use game theory to program the UAVs. Game theory, popularized by the 2001 film “A Beautiful Mind” about the economist John Nash, calls for individuals to act in their own best interests by taking into account what others will do.


“By using game theory, we are able to predict what other UAVs would do in their own circumstances,” Uyar said. “Each UAV is rational, selfish, and seeking the best outcome for itself. By predicting what the others will do, each UAV knows how to play a game.”


If this sounds like a lot of programming for a tiny vehicle, Uyar and Kusyk say that is where their research has advantages. “That’s why ours are better than other commercial ones because we know how to make these things computationally lightweight, with simple programs that run fast,” Uyar said.


Computer emulations have shown the software works, Uyar said. They reported some of their work at the 2019 Military Communications for the 21st Century (MILCOM) conference in Norfolk, Virginia.


“Our approach differs from existing security measure implementations reported in the literature since it synergistically combines evolutionary computation with game theory and stochastic processes to provide a powerful and comprehensive cyber defense system,” they wrote in their presentation to the conference. “In our approach, evolutionary methods using context-free grammars generate programs that detect and prevent real-life cyber threats,” they added.


The team is still working with the grant. “Currently, we are putting these programs into real UAVs but we haven’t flown a real UAV swarm,” Uyar said. “We need more funding.”