A termite mound is a model of insect engineering. Some are meters high and consist of a complex network of tunnels. Even more impressive, millions of the bugs work together to build the mound, all without a blueprint or foreman telling them what to do. Could robots do the same?
That’s a question that has now been tackled by Justin Werfel, a computer scientist at Harvard University Today, he and his colleagues introduced a computer program that figures out how autonomous robots can make specific structures, including small-scale skyscrapers and pyramids, simply by following the same set of rules. The researchers started small, tasking three compact robots, or bots, with making a one-story, three-pronged structure all on their own, a job they completed in 30 minutes.
The structure “is a proof of principle, a first step,” says Laurent Keller, an evolutionary biologist who uses robots to study social behavior in insects at the University of Lausanne in Switzerland and who was not involved in the work. “It would be interesting to see something [built] that is more complex.”
It’s a long-awaited first step, however. In 1995, researchers had proposed that such a feat might be possible , says Dario Floreano, a roboticist at the Swiss Federal Institute of Technology in Lausanne who was not involved in the work. “But it took almost 20 years before this could be realized—with [this] set of mobile robots.”
Werfel and his colleagues envision autonomous robots building sandbag barriers to prevent flooding or doing other types of hazardous work that people either couldn’t or would prefer not to do themselves. Their approach makes possible the use of many simple robots, any of which is expendable, instead of relying on a few sophisticated ones, where the loss of one might jeopardize a project.
To get robots to build specific structures, Werfel thought about how termites use a limited repertoire of behaviors to build their mounds. He developed a computer program that works backward from the desired structure, breaking its construction into a series of rules that the robots follow about which way to move and where to put down bricks. The order of the bricklaying is not specified, nor do the robots know what they’ve accomplished. Instead, the robots adjust what they are doing based on the presence of other robots and bricks already in their paths , Werfel and his colleagues report online today in Science and at the annual meeting of AAAS, Science’s publisher, in Chicago, Illinois. Like their termite models, the robots “don't have explicit instructions to do specific things,” Werfel explains. “Each is just reacting to what it encounters.”
To test whether this computer program and this decentralized approach to construction work, Harvard engineer Kirstin Petersen built liter carton-sized robots named Isis, Kali, and Nargun, equipped with four wheel-legs, called whegs, and a front-loader that could pick up and hold 21.5-square-centimeter foam bricks about 4 centimeters high. Sensors enable the bots to detect a black-and-white grid pattern on the bricks to help them navigate, and to sense the presence of the structure’s perimeter and other robots. The robots go forward, backward, turn, and climb up and down one step at a time.
Petersen programmed the three robots with the rules needed to make a 10-brick trident (see video). The researchers kicked off the process by laying down three bricks in a row and putting a white arrow on the floor next to these bricks. Each robot circled the bricks until it found the arrow, which prompted it to climb onto the starting bricks, turn to pick up a brick, then move and turn so it could place its brick on the base, and later, on one of the growing prongs of the trident. Then it climbed down and circled until it once again found the arrow to repeat the cycle. The robots don’t know when the structure is done—only when there are no more places to put bricks—and they keep circling until they run out of battery or are turned off, Petersen says.
The work “shows that with a simple behavior which was programmed into the robots, you can get a type of construction done,” Keller says. “From their models, they show that using similar rules, you can have more complex structures that can be built.”
The trident is not very complex, but as Barbara Webb, a bioroboticist at the University of Edinburgh in the United Kingdom who was not involved with the work, points out, “it’s always really hard to make these things work in the real world.” Indeed, Werfel says getting the bots up to speed for the job was quite the challenge, as robots tend to make small errors—they drift off course and can fall off the structure, they fail to pick up the brick, and so on—that add up and need to be corrected.
Nonetheless, Floreano says, “the paper shows that robots can be used for construction without a global supervision, making the concept applicable to environments where humans cannot go.”