Aerial additive manufacturing: drones to construct remote infrastructure (engineering.cmu.edu)

Carnegie Mellon researchers led by Amir Barati Farimani demonstrated an “aerial additive manufacturing” system that pairs AI-guided drones with modular magnetic blocks and a large language model (LLM) to assemble structures in flight. Rather than attempting fragile layer‑by‑layer 3D printing from a moving platform, the team uses drones for precise pick‑and‑place assembly, a top‑down camera for visual feedback, and an LLM that converts high‑level commands (“build a bridge”) into executable, adaptive plans. In lab tests on a 5×5 grid the closed feedback loop—camera detects placement errors, LLM replans instead of restarting—yielded a 90% construction success rate. For the AI/ML community this work is significant because it showcases LLMs functioning not just as language tools but as real‑time planners and error‑recovering controllers in robotic systems, integrating perception, symbolic planning and low‑level actuation. The approach highlights practical strategies for robustness (modular parts, replanning) and points toward multi‑agent coordination, in‑field adaptation, and domain transfer challenges—outdoor dynamics, richer 3D geometries and novel materials. Next steps include outdoor trials, extending LLMs to full 3D structure synthesis, and experimenting with dynamic building materials, all of which will push research on grounded LLM planning, sim‑to‑real transfer, and autonomous disaster‑response robotics.