A team of researchers at Harvard has figured out how to get a swarm of 1024 tiny robots to work together and self-assemble into complex two dimensional shapes without anyone interfering.
Called Kilobots, the robots’ behavior is intended to replicate fish forming into schools, birds flying in formation, and that sort of thing (a murder of Kilobots?). The researchers’ achievement—overcoming hardware and software hurdles that have previously limited such robo-swarm coordination to about 100 machines—is a significant step in developing collective artificial intelligence.
Videos by VICE
![](https://images.vice.com/motherboard/content-images/contentimage/14529/1408051822353291.png)
Robots assembled into a ‘K.’ Image: Michael Rubenstein, Harvard University
“Increasingly, we’re going to see large numbers of robots working together, whether its hundreds of robots cooperating to achieve environmental cleanup or a quick disaster response, or millions of self-driving cars on our highways,” said Radhika Nagpal, one of the researchers who worked on the technology. “Understanding how to design ‘good’ systems at that scale will be critical.”
To get the 1024 Kilobots to form a shape—which took about 12 hours—four robots mark the beginning of the swarm’s coordinate system, then the researchers uploaded a 2D image to the rest of the swarm. Using what the scientists describe as “primitive behaviors,” the Kilobots take turns moving into position, following the edge of a group. The whole time, they track their distance from ground zero, where the original four robots are located. When traffic jams happen or robots go off course, the swarm works together to solve the problem.
The Kilobots themselves are simple creatures. Moving around, sliding really, with vibration motors mounted atop three rigid legs, they use infrared receivers and transmitters to understand what’s around them and communicate with their robot collaborators.
The simplicity was necessary in order to keep costs down.
“In designing a large-scale robot swarm, the extent to which the robots can be fully autonomous (capable of computation, locomotion, sensing, and communication) must be balanced against the cost per robot,” Michael Rubenstein, one researchers wrote in Science, where the research was published. “Mass production favors robots with fewer and cheaper components, resulting in lower cost but also reduced capabilities and reliability.”
The algorithm, which is the real achievement of this research, consists of three major components: edge-following for movement, gradient formation, which helps a Kilobot understand how deep it is within the swarm, and localization, which helps each Kilobot establish where it is in the swarm’s collective coordinate system. Working together, the three components make it possible for the Kilobots to self-assemble into shapes.
The Kilobot swarm can’t form perfect shapes yet, as is obvious from the photographs. But that’s not the point. In the paper Rubenstein explains that even though no two shapes would look exactly the same, the degree of accuracy was consistent.
“The packing pattern of robots can exhibit considerable variability, as is true in natural self-assemblages,” he wrote, “These patterns are a manifestation of the natural variation in agent behavior in such large groups.”
Forming 2D shapes in a lab is just the beginning. As Nagpal hinted at, as robots become an increasing part of our lives, it’s going to be necessary to give them collaborative artificial intelligence if we want them to help build bridges, and clean up ocean garbage. And the hardware and software in Kilobots is a significant step toward that future.