scorecardresearch Skip to main content

At Harvard, tiny robots ‘swarm’ into shape

CAMBRIDGE -- When Harvard scientist Michael Rubenstein walks into the laboratory in the morning, he is greeted with a scene somewhere between a disco and the opening of a science fiction movie about a robot apocalypse. A constellation of LED lights blinks in the darkness -- the electronic heartbeat of his 1,024-robot horde. They are ready to do his bidding.

For years, engineers have talked about the future of robotics: not one polished machine with all the dexterity and intelligence of a human being, but a swarm of individually dumb robots that could work together to accomplish big tasks. Those could range from building a base on another planet before human explorers arrive, to crawling into rubble after an earthquake to search for survivors. But researchers’ “swarms” have tended to peak at 100.

Now, with their miniature robot army, three Harvard University researchers have upped the ante, assembling a massive swarm of simple, three-legged robots that can work as team. The robots -- called Kilobots -- can crowd together like a mechanical school of fish to form specific shapes, such as a star or the letter K. They can synchronize their clocks like fireflies, or move toward a light source like a plant.


The advance, reported Thursday in the journal Science, is a feat of “engineering majesty,” said James McLurkin, director of the Multi-Robot Systems Lab at Rice University, who was not involved in the research.

“Building 1,000 robots is hard,” McLurkin said. “Getting 1,000 robots to work together reliably is, how they’d say it in Boston? ‘Wicked hard.’ ”

The technology is still in the early stages. These simple robots, which each weigh about as much as three nickels and cost $14 in parts, can’t build a skyscraper or clean up an oil spill. But they surmount several major problems in robotics, McLurkin said.

The software the researchers designed allows the individual robots to act on their own, using only information from their neighbors to achieve goals that dwarf their thumb-sized bodies. That means they aren’t following the master plan of a central brain directing them all what to do. To build a particular shape, they have to follow the edge of the group and measure their distance from a marker robot.


The experiment also reveals the limits of computer simulations that have been used to project how a swarm of robots would behave. Such models miss a lot of the real-life challenges, such as what happens when individual robots jostle each other, stop working, or just move slower than the others.

“When you make simulations, you’re always simplifying the world,” said Radhika Nagpal, a professor of computer science at Harvard and the Wyss Institute for Biologically Inspired Engineering, who oversaw the study. “We realized there are some errors we never even considered putting into simulations” when robots are interacting in the real world.

The researchers also dealt with logistical challenges so trivial that they only become challenges when you’re building 1,000 robots, not six. The lab held robot-assembly parties, which consisted mostly of pulling stickers off the little buzzing motors -- the same ones that are used to make cellphones vibrate -- and removing batteries from packaging. Merely charging and activating 1,024 robots is no small feat, so they designed a table where the robots could be wrangled onto a metal plate to be charged. Rubenstein herds them by pushing them with a yardstick when they are low on batteries.

The researchers are thinking about ways to make the robots even smaller and simpler, with the idea that like cells that form an organism, they may be able to assemble into bigger, more complicated objects. In one demonstration, researchers had the robots move around until they were in the shape of a wrench -- one day, that approach might be a way to assemble useful objects on demand.


Pei Zhang, associate research professor at Carnegie Mellon University, said that although there is no immediate application for this group of robots, it was an important step toward even more miniature robots.

“This work is a step to building tiny dust-size robots,” Zhang wrote in an e-mail. The possibilities are limitless, he said, from self-healing space ships to a surgical tool that could shape itself to cut out only cancer cells.

What may be most instructive, however, is what goes wrong as the researchers tweak their army. During a recent demonstration in which robots were supposed to head toward the light, one started moving around in befuddled circles. Given that some robots will always fail, they have begun to work on ways for those robots to be alerted that they aren’t working and reboot.

“These are the issues that I think are really new in a sense,” said Francesco Mondada, a robotics engineer at École Polytechnique Fédérale de Lausanne in Switzerland. Although researchers have claimed to be working on swarms in the past, their swarms have often amounted to a handful. This, he said, is unequivocally a swarm.

Carolyn Y. Johnson can be reached at Follow her on Twitter @carolynyjohnson.