assembly 3D model / panels breakdown
In this project we’ve explored neural networks that work with 3D objects in order to generate new forms and interpret them through physical matter. We've maximized automation of digital to physical process by using algorithms that split the original form into discrete parts and used robotic arms for assembly. As a result we've received a 1m x 1m assembled object.
Process:
We’ve constructed a dataset containing 3D architectural elements and trained a 3D neural network to predict new forms. After studying the results we continued working with one object that was dissected into panels using python scripts.
The structure is fully self-supported and composed out of repetition of 2 sizes of wooden blocks. Robotic assembly line was designed to pick the corresponding block and place it in a specified location in order to complete the panels.
The texture was generated using CycleGan neural network trained on Moebius illustrations. Results were digitally mapped on the panels; printed and applied to the final assembly.
dataset input examples
dataset output examples
panel texture
panel texture
panel texture
panel texture
assembly process
