Disaster just struck, roads are inaccessible, and people need shelter now. Rather than wait days for a rescue team, a fleet of AI-guided drones takes flight carrying materials and the ability to build shelter, reinforce infrastructure, and construct bridges to reconnect people with safety.

It sounds like science fiction, but new research from Carnegie Mellon University’s College of Engineering combines drones, additive manufacturing, and large language models to rethink the future of aerial construction.

\

Aerial additive manufacturing (AM)—think flying 3D printers, has been fascinating researchers for years, but the natural instability of a drone in flight makes traditional layer-by-layer fabrication nearly impossible. To overcome this, Amir Barati Farimani, associate professor of mechanical engineering, has equipped drones with magnetic blocks to allow for precise pick-and-place assembly and a large language model (LLM) that can translate high-level design goals like “build a bridge” into executable plans.

“The adaptability of LLMs allows us to generate and adapt building plans onsite. If we encounter problems while building, we can switch gears to ensure efficient and accurate construction,” said Barati Farimani.

Drones equipped with magnetic blocks have precise pick-and-place assembly. A large language model can translate high-level design goals like “build a bridge” into executable plans to be used with such drones.

To test this take on aerial AM, researchers set up a 5x5 grid and tasked drones with designing specific shapes using the magnetic blocks. Because the drones were monitored by a camera, if they dropped a block in the wrong position, left a gap, or built inefficiently, the LLM autonomously prompted the drone to work with the error from a new plan rather than starting over. Thanks to this closed feedback loop, construction builds were successful 90% of the time.

We can imagine this tool filling potholes, fixing spaceships in orbit, and constructing infrastructure in mountainous regions where heavy machinery can’t reach.

Amir Barati Farimani, Associate Professor, Mechanical Engineering

“We can imagine this tool filling potholes, fixing spaceships in orbit, and constructing infrastructure in mountainous regions where heavy machinery can’t reach,” he said.

Moving forward, the team plans to test their drones outside of the lab to address future real-world challenges. They plan to explore using LLMs to construct 3D structures, and to work with more dynamic building materials that would further optimize the performance flexibility of construction designs.