VAULT AI – Computational design and fabrication of self-supporting VAULTed structures through AI-driven human-robot cooperation 

VAULT-AI investigates AI-supported methods for the design and in robotic situ fabrication of self-supporting vaulted structures. The project integrates form-finding algorithms, structural optimisation, and human-robot cooperative assembly to achieve material-efficient, waste-free vaulted constructions that eliminate the need for scaffolding or falsework.

Masonry vaults use their distinctive double curvature to carry loads with very little material. Structural design often focuses on the final, completed structure, while assembly of large-span vaults typically requires extensive temporary support, adding material waste, cost, and construction time. VAULT-AI aims to advance the real-world deployment of vaulted construction by integrating fabrication-aware structural form-finding with 3D perception and spatial AI, enabling vault geometries that are not only structurally and materially efficient, but can also be assembled as self-supporting structures through seamless collaboration between mobile robots and human builders. AI-based action recognition further supports safe robotic manipulation during construction.

VAULT-AI is thus structured around four interconnected work packages: fabrication-aware structural form-finding, on-site cooperative robotic fabrication, spatial AI for human-robot perception, and machine learning for mobile robotic control. We aim at validating the effectiveness of our methods through experimental case studies featuring self-supporting masonry vaults, demonstrating how an integrated AI-driven human-robot system can enhance material efficiency, flexibility, and safety in construction.

 Collaborators:

• Likhinya KVS, M.Sc., Professorship of Digital Fabrication (Prof. Dr. Kathrin Dörfler), TUM
• Jesus Daniel Meza Zeron, M.Sc., Professorship of Structural Design (Prof. Dr. Pierluigi D'Acunto), TUM
• Niklas Schlüter, M.Sc., Chair of Security, Performance and Reliability for Learning Systems, School of Computation, Information and Technology (Prof. Dr. Angela P. Schoellig), TUM
• Anran Zhang & Simon Boche, M. Sc., Previously: Professorship of Machine Learning for Robotics, School of Computation, Information and Technology, Now: Mobile Robotics Lab, ETH Zürich (Prof. Dr. Stefan Leutenegger); ETHZ

 Funding programme:

This research is funded by the TUM Georg Nemetschek Institute Artificial Intelligence for the Built World (TUM GNI)

 Duration:

• 2025-2029