Abstract

Providing an online interface for people to generate their own custom furniture. After picking the type of furniture needed, the Neural Network would direct the user towards a solution using uploaded images of what the user likes (photographs, painting, sculptures…). The website would then allow the user to export building plans for his chosen design, or have the furniture made by a professional. The “design spaces” (parametric models) are made by different designers that can upload their work.

AI-human cooperative design

Using AI for design customization?

Frequently in the creative process when starting a new project, the first step is to talk with the client to understand the needs and possible variants in which the project can be developed.
This process can become quite tedious if the ideas are opposed or not related correctly; To make this process more efficient, we decided to develop a tool that allows the designer to visualize different configurations of his client’s ideas through his own visual references combined with an algorithm that combines the input and gets analyzed by a neural network.

  • Human input
  • Evaluation

To carry out this project, we decided to develop a questionnaire to collect the consideration of characteristics given by different people of the relationship between selected images and designs of different chairs.

Rietveld chair

Inspiration

Parametric chair model first iteration

Parametric chair: final model

Surveys

Image/Design Traits

Scoring system

This is the method that we used to score the different attributes given by the person

Images

Design

Image/Design

Survey: results

Images

Images with clearest distinction

Designs

Designs with clearest distinction

Creating the Activation Network

Basic setup

Training tests

This are the different configurations we used to train the Neural Network

Constant layers

Constant parameters

Networked used for results

Visual comparison

Networked used for results

Network inputs

Different ways of using the network

NN designs

Manipulating the traits

Using one image

Using multiple images

Human UI 

Using the plug-in for grasshopper Human (by andheum) we developed an interface so the user can have the instant changes in the configuration of the chair, regarding the input given

Autoencoder

Dimension reduction

Platform

Concept

Home

Finally an exploration of how this project could look like taken in to the market

Select Furniture Typology

Input images

 

Export building plans

References:

Gerrit Rietveld Red Blue Chair 1918–1923

Survey Images:

Muralla Roja by Ricardo Bofill, 50 Manga chairs by Nendo, Interior of the Pantheon by Giovanni Paolo Panini, Grey Abstract Sculpture by unknown, Route de l’estaque by Georges Braque, Black square by Kazimir Malevich, Number 32 by Jackson Pollock, Psychedelic art  by unknown, David by Michelangelo, Drawing by Ana Kras, Sculpture 21c by Karl Andreas Groß, Untitled by Fujikasa Satoko, Etude pour mercure by  pablo picasso, Soviet bus stop, Mechanical Elements by Fernand Léger, Minimal art by unknown, Photo by Nicholas Alan Cope, Nude descending a staircase  by Marcel Duchamp, Salón de Gasparini hall in the Royal Palace of Madrid, Neurons  by Santiago Ramon y Cajal, Detail of Fresco painting in St. Charles’s church in Vienna, Ceiling detail by Antoni Gaudi, Labyrismen by Constant Nieuwenhuys, Mobiles by Alexander Calder, Mask 5, Series 1 by Neri Oxman

Useful links:

Owl(by Mateusz Zwierzycki)

Qualtrics.XM

Human (by andheum)

Faculty:

Mateusz Zwierzycki

Soroush Garivani

Students:

Lorenzo Masini

Cedric Droogmans

Elena Jaramazovic

Luis Jayme Buerba