Generative Physical Objects: Text to Physical Objects using
3D Generative AI, Vision Language Models, and Robotic Assembly
Name
kyaw-alexkyaw-smarchs-smeecs-arch-eecs-2026-thesis.pdf
Size
22.4 MB
Format
Adobe PDF
Checksum (MD5)
31cc799bb0aebdc99847e5d30f3ce501
Author(s)
Kyaw, Alexander Htet
Advisor(s)
Sass, Lawrence
Davis, Randall
Date Issued
February 2026
Publisher
Massachusetts Institute of Technology
Abstract
The thesis presents a framework for transforming natural-language input into physical objects using 3D generative AI and robotic assembly. While 3D generative AI lowers the barrier to creating digital geometry, translating those outputs into physical objects is constrained by real-world limitations. Therefore, the thesis formalizes five key considerations for making physical objects with generative AI: accessibility, sustainability, time, fabrication constraints, and functionality. Guided by these considerations, the thesis presents an end-to-end workflow connecting 3D generative AI with discrete robotic assembly by integrating mesh discretization, geometric processing, and robotic path planning. This includes identifying a set of fabrication constraints to ensure the AI‑generated geometry is feasible, including component count, overhang stability, vertical stacking, connectivity, and robot reachability. The thesis further expands the system to support complex multi‑component objects by using a vision‑language model for function and geometry-aware component assignment. To support user modifications, the system incorporates a human-in-the-loop workflow that allows users to provide feedback and steer the assembly outcome. Results demonstrate that the system can construct a range of multi-component objects within minutes and support user refinement, aligning with the speed and variability of 3D generative models. Finally, the thesis presents alternative variations for the on-demand production pipeline, exploring new possibilities with different component geometries, mobile robots, and gestural input modalities. Together, these contributions establish a framework for human-AI co-creation that connects natural language, generative AI, and physical making.
MIT Department
Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science
Massachusetts Institute of Technology. Department of Architecture
Terms of Use
Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0)
Copyright retained by author(s)
Persistent DSpace Link
