How Do Drone Light Shows Work?
USC’s Multi-Robot Systems conference probes the science behind Borg-like UAVs.
If you were in Los Angeles on September 14 and looked up into the night sky, you would have seen a swarm of 300 Intel Shooting Star drones, equipped with LED lights andflying in sequence to form a huge W.
Despite initial paranoia in downtown that the display was a subliminal message from a more advanced civilization, the drones were in fact working for Warner Bros/DC Comics, promoting the home entertainment release of Wonder Woman.
Cool studio marketing stunts aside, you might have been curious to learn how drones fly in formation. That’s what PCMag wanted to know, so we went to the first Multi-Robot Systems (MRS) conference at University of Southern California (USC) to grasp the science behind Borg-like UAVs.
MRS is a new initiative of the IEEE RAS Technical Committee on Multi-Robot Systems, and it’s intended to bring together researchers who are in the field of multi-robot systems (MRS) and multi-agent systems (MAS). “Typically MRS/MAS research is spread across large conferences, so the intent of this more focused conference is to bring those researchers together to highlight the best in the field and learn more from each other,” said Dr. Nora Ayanian, Assistant Professor and Director of the ACT Lab at USC and MRS chair.
It was definitely an advanced academic symposium, with peer-reviewed papers presented in quick fire rounds. But the scientific principles were fascinating.
Most computer science programming is based on “if/then” — essentially procedural decision making. But in order to have true automation and AI, drones need to get modular, become self-aware, and make decisions based on experience, (i.e. like humans) in their style of problem solving.
Dr. Ayanian joined the Viterbi School of Engineering at USC’s Department of Computer Science in 2013, after doing post-doc work at MIT’s Distributed Robotics Lab (CSAIL). Her work focuses on end-to-end solutions for multirobot coordination, which start from truly high-level specifications and deliver code for individual robots in the system.
“Currently, multirobot systems are extremely difficult to use,” Dr. Ayanian explained. “And solutions are not transferable from one application to another. So the goal of my research is to develop distributed planning and control foundations that are broadly applicable across all aspects of multirobot systems or mobile sensor networks.”
If you want to see her work on quadrotor UAV swarms in action, check out the video below.
At the symposium, Shauharda Khadka, a graduate research assistant at Oregon State University currently pursuing a PhD in Robotics, showed how his autonomous units (nicknamed WALL-E and EVE, naturally), learned to navigate an environment to track rewards that changed location rapidly and unpredictably. The team of robots learned to explore an environment to track a reward, exploit the reward when they find it, and resume exploration when its location changed.
Simply put, UAVs developed a “collective mind” in order to swarm in formation, memorize salient observations, and adapt their policies rapidly to new stimuli in their surroundings.
UAVs are often deployed in places where first responders struggle to gain access, like disaster zones. Dr. Michael W. Otte, a Postdoctoral Associate at the U.S. Naval Research Laboratory (NRL) who will soon join the Department of Aerospace Engineering at the University of Maryland, demonstrated how his algorithms — which draw on distributed systems, graph theory, probability theory, and auctions — enable robotic teams to pool resources in order to solve their common problems.
The UAVs were able to wirelessly “bid for tasks” in an auction-style environment. That is, you could set off a group of drones in the direction of a disaster area and, on arrival, they could break off to undertake certain much-needed tasks.
“Multi-robot teams often face problems that require dividing a set of tasks among the team’s robots, so a popular way to allocate tasks [among robots] is with an auction,” Dr. Otte said. “Items are sold to the highest or lowest bidder, and we study how communication quality affects the results of the auction.”
In other words, can auctions work if many of the bidders can’t send in their bid? Research indicates that that, yes, it turns out they often still work pretty well. So, when you next see UAVs in formation, there’s not just an enormous amount of aerodynamics keeping them in the air, but smart algorithms enabling purely autonomous behavior within a constantly changing group. Now they just have to build in silencers because there’s nothing stealth about a single drone, let alone 300 in formation.
If you’ve been stockpiling drones and want to get into the swarming field, the FAA has released a certification for remote pilots. You’ll need an exemption to traverse air space in your geographical location, and a deep understanding of multi-agent control mechanisms to provide self-awareness in your devices, of course.
Read more: “Trust Me, I’m a Robot”
Originally published at www.pcmag.com.
