TERMES Project Models Robots Based on Collective Intelligence of Termites

Though typically associated with the destruction of structures, termite colonies may have inspired the next big innovation in construction. A team of engineers and computer scientists at the School of Engineering and Applied Sciences and the Wyss Institute for Biologically Inspired Engineering built a “colony” of autonomous, interchangeable robots—coined the TERMES project—based on the construction strategies of termites and other insect species, according to a report published in Science earlier this month.

The TERMES report, co-authored by computer science professor Radhika Nagpal and graduate students and researchers Justin K. Werfel and Kirstin H. Petersen, is founded on the notion of collective intelligence. Trained as an engineer and a physicist, Werfel began collaborating with Nagpal while he was a Ph.D. candidate. Together, the two developed an interest in termites’ mound-building techniques.

“There are a number of insects that build,” Werfel said. “Termites inspired us by the scale of what they build. They’re less than one centimeter long, and they build the equivalent of skyscrapers.”

Over the course of researching mound-building termites, Werfel came across an article Petersen had published regarding a similar interest in the insects, but with an emphasis on hardware. Reaching out to Petersen, Werfel discovered that their studies aligned in “a very complementary way.”

“My question was how to program these robots, and hers was how would we build them,” Werfel said.

Unlike centralized intelligence systems, the TERMES robots are interchangeable, independent, and expendable. Each individual robot is equipped with four types of sensors that allow it to navigate its local environment. These robots are not assigned specific roles. While this model of collective intelligence sacrifices the flexibility and efficiency of a centralized system, the decentralized system compensates in other ways.

“One of the advantages of the decentralized system is that it is more robust to losing individual components,” Werfel said. “If some of the robots break or are lost, it doesn’t matter to any of the others.”

The researchers suggest this model of a construction crew—which requires no supervision and boasts low costs once the robots are built and programmed—opens up possibilities for human construction projects as well. Instead of delineating tasks to each worker, an architect could present an outcome to the robots, and they could be programmed to carry out the task automatically.

Spearheaded by engineers, the project has led its researchers into the fields of biology and architecture. All three authors have visited Namibian termite hordes to implement data collection mechanisms in native mound-building colonies.

“Obviously we’re not entomologists, and we’re not trying to do something they can do much better,” Petersen said. “But we’re computer scientists, so we can help them by building tools that will allow them to gather much more quantitative and qualitative data on the termites.”

Underlying Nagpal, Werfel, and Petersen’s work on the termite-inspired robots are a series of broader questions about the emergent outcomes of many individual actions.

“It’s easy if you have a lot of low-level rules to predict the high-level outcome,” Petersen said. “If you have specific high-level outcome in mind, it’s hard to figure out what the lower-level rules should be.”

—Staff writer Jessica A. Barzilay can be reached at jessica.barzilay@thecrimson.com. Follow her on Twitter @jessicabarzilay.