Karen Asmar
Karen Asmar
Architect - Creative Technologist - Designer

 

 

This project is hosted on my older website.
For my latest (still-in-progress) work, visit www.karenasmar.com.

 

SIGraDi 2020 / MFA Thesis Spatial Design Tools

Machinic Interpolations

A GAN Pipeline for Integrating Lateral Thinking in Computational Tools of Architecture

 
1-Cover (1).gif
 

CHALLENGE

Machinic Interpolation targets the need to integrate lateral thinking strategies in digital design tools. Many of the tools used by architects to design, conceptualize and experiment have entered into the discipline from fields such as engineering, manufacturing or animation. As a result, values such as optimization, standardization and efficiency have discretely found their way into these tools and have greatly informed and constrained the possible design space.


SOLUTION

Within this context, the project proposes a methodology that uses GANs (Generative Adversarial Networks) and their properties as an experimentation ground to reevaluate the lateral thinking constraints in architectural tools. Specifically, it uses StyleGAN and explores the ability to access its latent space as part of an architectural design process.

The presented methodology constitutes a 4-step approach that draws from the abilities and properties of this space to design architecture: initializing the virtual environment of points (through training), entering the space (randomly or with an intent), moving through space (through arithmetic operations and interpolations), and finally generating voxelized 3D forms from interpolated images.

By doing this, the project reveals how this series of techniques could lead to unexpected spaces, resulting in creative output beyond what is produced by human capability alone.

 

CONTEXT

MFA Thesis Project
Parsons School of Design
2020

In May 2020, I graduated from Parsons School of Design where I pursued a Masters in Design & Technology on a President's Scholarship. My coursework and projects focused on exploring new ways of interacting with, experiencing and designing the built environment through emerging technologies.

This project presents my master's thesis, advised by Harpreet Sareen. The recorded talk can be found here.

PUBLICATION

SIGraDi 2020 Conference
Transformative Design

TAGS

Critical Thinking,
Data Collection,
Machine Learning,
Generative Adversarial
Networks (GANs),
Creative Coding,
UX, UI Tool Design

 
process.jpg
 
 
 

Thesis Presentation —

Recorded Talk01*

 
 
 
 
 

Problem Space —

Research01*

 
 

Evolution of Tools

Surfaces on which we draw and experiment have greatly changed over the years. Architects once carved on stone and pushed ink and graphite across papers. Today, architects move cursors and push keys and buttons to design on digital screens.

These screen-based interfaces we design with today are far more complex than the orthographic surfaces and the pen and paper we are used to. What appears on those screens behaves in ways we are not as familiar with and in reality forms a very small fragment of what is really happening wihtin those machines.

 
evo.png
 
 

Replicating Orthographic System

The digital environments we design in today tend to replicate the orthographic system embedded in our discipline and that we inherit from generation to generation.

Yet, the machines we design on do not function the same way. These machines work with pixels and speak a language of numbers. Machines allow us for larger array of possibilities other than mechanical behaviors and limitations of physical material such as split, trim, chamfer etc.

 
pixel_03.jpg
 

Pen & Paper

Interfaces & Machines

 
 
 
 

Yet…

The task of any architect is not about using computers to replicate or to automatize what has already been thought of and produced, it is about allowing computers to reveal a fully new form of architectural intelligence. Machines, [...] shall logically give birth to a kind of architecture that is also beyond our usual capacities. Such an architecture is still to be produced.
— Philippe Morel
 
 

Hence,

Instead of replicating what has already been achieved, how can we use machinic techniques to imagine new computational tools of architecture?

 
 
 
probmap.png

Why is this Important?

 
 

Dom-ino system, 1914. Le Corbusier

 

But why do we need new ways of designing?

“I find it incredibly odd that our world has changed but our vision of the built environment hasn’t”
- Mollie Claypool

 
 

Define —

Opportunity03*

 
 
 
 
download.jpeg
 
 
LatentSpace.jpg
 
 
 

TECH REPORT


 
 

How —

Methodology*

 
 
pipelineflow-7083.gif
 
01_LS.gif
 
 

Latent Space

A latent space is an n-dimensional space where, after training a GAN model, the generator of this model learns to map points in this space to specific generated images, similar to the ones of the dataset it was initially trained on.

 
 
 
 

Four-Step Process

The 4-step process draws from the abilities in and properties of the latent space of the machine to design a 3D structure. This figure showcases this pipeline and shows how it can push architectural digital tools into the grounds of experimentation.

labellllsss.jpg
 
latent.gif
 

Design Space

The latent space has a structure that can be queried and navigated where every point in this space is a 512D vector that represents an image.

The virtual environment represents the digital design space in which architects can design, analyze, edit and generate 3D structures. The current tools used, such as Rhino3D, focus on orthographic approaches and rely on curves, lines, surfaces, NURBS, etc. and the manipulation of those elements to create designs in a 3D virtual space.

 
 
 

Technology Details —

How it Functions*

 
 
 

01

03_latentspace.gif
 
 
04_latentspace.001.jpeg
 

02

 
 

03

 
06_latentspace.gif

04

 
 
 
07_latentspace.001.jpg
interpolation3D.gif
 
 
 

Experimenting with Possibilities —

Results of
Tool Pipeline*

 
 
 

01

 

02

 
mix-2000x1344.jpg
 
 

03

 

04

 
changing.gif

Changing Directions

 
 
 

But, many options exist...

optionss.gif
 
 
 
6-3dExtrasGrid-7628.jpg
 
black.jpg