Watch An Autonomous Robot Swarm Form 2D Starfishes

The first thousand-robot flash mob has assembled at Harvard University.

by Beckett Mufson

August 14, 2014, 5:45pm

Images courtesy of Michael Rubenstein and Science/AAAS

Researchers have surpassed a milestone in the field of collective artificial intelligence, and it's taking the shape of… a starfish. What separates this latest development from its predecessors is that the new, experimental 'intelligent' robots—dubbed Kilobots—took this form completely autonomously, like a swarm of bees gathering around their hive.

Developed by the Harvard School Of Engineering And Applied Science, each Kilobot is simply structured, consisting of three rigid legs, a body just a few centimeters in width, an infrared transmitter and receiver, and a small, vibrating motor. The Kilobots communicate with one another through infrared light beams, sensing their proximity to one another, but lack a bird's eye view to align themselves by in real time, which is how the Pixelbots we reported on in April maintained their shape. “These robots are much simpler than many conventional robots, and as a result, their abilities are more variable and less reliable,” said lead author Michael Rubenstein in a press release. “For example, the Kilobots have trouble moving in a straight line, and the accuracy of distance sensing can vary from robot to robot.”

A major part of the breakthrough comes from an algorithm Rubenstein and his research team developed to circumvent the flaws of the exceptionally simple robots. Under the power of the Harvard team's coded instructions, the Kilobots can work together to correct traffic jams and incorrect positioning. Using the infrared communicators, they work outward from four designated origin point bots, conforming to a simple 2D image. They use simple functions—i.e. following the edge of a group, tracking their distance from the origin bots, and maintaining a sense of relative location, herd behavior-style—to slowly troubleshoot their way into forming the desired shape.

“At some level you no longer even see the individuals; you just see the collective as an entity to itself,” says Radhika Nagpal, a core member of the Wyss Institute for Biological Engineering, which supported the Kilobots research. Nagpal was also involved in the TERMES construction robots, a predecessor to the Kilobots that demonstrated similar principles—collaborative AI working together to build something without human supervision. However, before the development of the Kilobots, no distributed robot collective had successfully exceeded much more than 100 individual robots. As suggested by their name, Kilobots can successfully follow instructions in an enormous, thousand-bot swarm.

The robot swarm could be used in the future as a platform to test increasingly complex, collaborative AI algorithms in the real world, rather than in imperfect simulations. "A simulation can only go so far,” says Nagpal. “The real-world dynamics—the physical interactions and variability—make a difference."

Kilobots still operate on an incredibly simplistic level (worry not, if you fear the impending robot apocalypse), but we hope to see them used for growth in scientific understanding of collective AI. In the meantime, we really just want to see our names written by an army of intelligent robots. Our Instagrams would be swarming with likes.