Teaching robots to build without blueprints

(Press-News.org) Bees, ants and termites don’t need blueprints. They may have queens, but none of these species breed architects or construction managers. Each insect worker, or drone, simply responds to cues like warmth or the presence or absence of building material. Unlike human manufacturing, the grand design emerges simply from the collective action of the drones — no central planning required.

Now, researchers at Penn Engineering have developed mathematical rules that allow virtual swarms of tiny robots to do the same. In computer simulations, the robots built honeycomb-like structures without ever following — or even being able to comprehend — a plan.

“Though what we have done is just a first step, it is a new strategy that could ultimately lead to a new paradigm in manufacturing,” says Jordan Raney, Associate Professor in Mechanical Engineering and Applied Mechanics (MEAM), and the co-senior author of a new paper in Science Advances. “Even 3D printers work step by step, resulting in what we call a brittle process. One simple mistake, like a clogged nozzle, ruins the entire process.”

Manufacturing using the team’s new strategy could prove more robust — no hive stops construction because a single bee makes a mistake — and adaptable, allowing for the construction of complex structures onsite rather than in a factory. “We’ve just scratched the surface,” says Raney. “We’re used to tools that execute a plan. Here, we’re asking: how does order emerge without one?”

 

A New Paradigm for Building  

From stone tools to space stations, human engineering has relied on planning: imagine the result, then design and build it in steps. Even 3D printing follows the same logic, slicing a model into thousands of precise instructions for the printer to follow.

“What’s so different about our approach is that it sidesteps that entire paradigm,” says Mark Yim, Asa Whitney Professor in MEAM, Ruzena Bajcsy Director of the General Robotics, Automation, Sensing and Perception (GRASP) Lab and the paper’s other co-senior author. “There’s no pre-written script, no centralized plan. Each robot just reacts to its surroundings.” 

Because no single robot needs to understand the big picture, construction can continue even if some robots fail or go off course. And since all robots operate simultaneously, rather than waiting their turn, the process could one day be faster — and more robust to individual failures.

 

Planning Behavior, Not Buildings  

While inspired by nature, the researchers didn’t try to precisely mimic how bees, ants or other natural builders behave. Unlike artificial intelligence researchers, who often look to the brain for clues about how to design learning algorithms, this team wasn’t trying to copy biology. 

Instead, they focused on the deeper principle that nature uses: simple behaviors, repeated many times in parallel, can add up to create something complex and useful.

“What we wanted was a system where structure emerges from behavior,” says Raney. “Not because the robots know what they’re building, but because they’re following the right set of local rules.”

The hard part was figuring out what those rules should be. “There are countless ways you could program a robot to react to its surroundings,” says Yim. “We had to narrow it down to something simple, but still useful.” 

 

Finding the Right Rules  

In the end, the team focused on a handful of basic questions: What should a robot do when it bumps into something another robot built? Should it turn left or right, and by how much? How far should each robot go before stopping?

This resulted in a dozen variables — like the robots’ speed and the angle at which they turn left or right — that the researchers played with over the course of many simulations. “By simulating the robots’ activity,” says Raney, “we could focus on fine-tuning which rules mattered the most.” 

Ultimately, the amount of disorder in the system played a crucial role in the final structure. “The more we varied parameters like the turning angle, the more variation we got in the final structure,” says Yim. 

As prior work by Penn Engineers has found, adding the right amount of disorder to lattices like honeycombs can actually enhance their toughness. “We essentially found a lever that lets you vary the geometry of the final outcome, which can affect its resistance to cracking,” adds Raney.

 

Building the Swarm in Reality  

While the team created prototypes, actually building a swarm of robots is still a step away. First, they plan to update their simulation to better reflect how tiny robots might work in the real world. 

“In our early models, we imagined the robots laying down material in straight lines, like a mini 3D printer,” says Yim. “But that may not be the most practical method. A better approach might be to use electrochemistry, where the robots grow metal structures around themselves.”

Making that happen will require progress in building tiny robots that can move, sense and interact with materials, but the team believes the concept itself represents perhaps their most important advance. 

“We hope this gets people thinking in new ways about how things can be built,” says Raney. “Nature doesn’t start with a master plan, it starts with lots of small actions that come together into something bigger. Now we’re learning how to do that, too.”

This study was conducted at the University of Pennsylvania School of Engineering and Applied Science and supported by the National Science Foundation (2036881).

Additional coauthors include co-first authors Jiakun Lu and Xiaoheng Zhu, as well as Walker Gosrich, all of Penn Engineering.

END