Final Presentation 29 Nov 2018
CHAPTER INDEX
INITIAL RESEARCH
What does it mean for AI to design?
Our initial research started with questioning the role of artificial intelligence in the field of design. What is it, and how is it used? We found:
- Artificial intelligence is the stacking of acute, specific, and specialized tasks to resemble complex human problem-solving processes. Results produced anteriorly may be used as input again in the same system - this produces a heuristic algorithm.
- The strength of AI lies in the speed of processing and ability for finding abstraction in systems.
- In contemporary design, AI is used to optimize (any parameter) to create structure or layouts, or to produce large quantities of variation on the same parameters.
- Publically available AI satisfies either of these two tasks: Optimized structure or quantities of variation.
Geometric Optimization by Evolutionary Algorithm
Evolving Floorplans by Joel Simon
Project Dreamcatcher
ParaGen by Peter Von Buelow
Graphic Design Generation
Logojoy
Nutella Unica
LOOK BOOK
Reflection, 2008 by Andreas Fischer & Benjamin Maus
Description- A music piece translated by using its notes to form a visual output.
Takeaway- Use a highly specific rule set to generate the design output.
Source- Generative Design: Visualize, Program, and Create with Processing by Hartmut Bohnacker, Benedikt Grob, Julia Laub
Recursive Growth, 2007 by Marc Fornes
Description- Using a recursive subdivision technique to offer the possibility of structural form finding.
Takeaway- Examination of growth processes resulting from recursive subdivision of surfaces.
Source- Generative Design: Visualize, Program, and Create with Processing by Hartmut Bohnacker, Benedikt Grob, Julia Laub
Aperiodic Symmetries, 2009 by Marc Fornes & Skylar Tibbits
Description- Complex spatial structures that are generated using optimization algorithms.
Takeaway- Optimization or the fitness function can range from aesthetic quality to cost, airflow, light transmission and structural optimization.
Source- Generative Design: Visualize, Program, and Create with Processing by Hartmut Bohnacker, Benedikt Grob, Julia Laub
Platonic Solids, 2008 by Michael Hansmeyer
Description- The project explores the use of three-dimensional subdivision algorithms and demonstrates that heterogeneous and rich forms can be generated using a single, simple approach on a platonic solid.
Takeaway- Changes made to the parameters directly affect form attributes such as branching, curvature, structure and surface properties.
Source- Generative Design: Visualize, Program, and Create with Processing by Hartmut Bohnacker, Benedikt Grob, Julia Laub
Delaunay Raster, 2008/2009 by Jonathan Puckey
Description- The Delaunay Raster provides feedback that allows the user to respond to changes in the image and influence the process directly.
Takeaway- The predictability of a more traditional triangulation filter is reduced by allowing for more user AI collaboration.
Source- Generative Design: Visualize, Program, and Create with Processing by Hartmut Bohnacker, Benedikt Grob, Julia Laub
Subdivided Pavilions, 2006 by Michael Hansmeyer
Description- Project examines 3 dimensional subdivision processes as generators of architectural forms.
Takeaway- In this work, source data consists of simple rectangular abstractions of a building. Space and contect determine the subdivision processes applied to a building.
Source- Generative Design: Visualize, Program, and Create with Processing by Hartmut Bohnacker, Benedikt Grob, Julia Laub
Other references:
Proposal for Collective Living II (Homage to Sir John Soane)
“If architecture organizes the world around us then making architecture from architecture is about reorganizing what exists in the world for new purposes”
Source - Proposal for Collective Living II (Homage to Sir John Soane) byANDREW KOVACS
Use of Genetic Algorithm to produce multiple iterations of clothing in Fashion Design
Studies on Fractal Developments
Source- ACADIA 2017, Disciplines Disruption
MAYA MODELS
10 Selected Designs
TEAM APPROACH - TECHNIQUES/SKILLS DEVELOPED
Volumetric Subdivision using an input curve.
Geometric Optimization by Evolutionary Algorithm
Galapagos Evolutionary Solver to generate an optimized curve.
LunchBoxML to spatially sort data based on similarities and differences
Image Sorting and Filtering
Text to Image (https://t2i.cvalenzuelab.com/)
Image Manipulation
Mobile Apps
Enlight Pixaloop
AI detects the background (sky) and make it in motion. Style also applied.
Prisma Photo Editor
Frozio Photo Animator
Pikazo - AI Art
Low Poly
The AI detects the edge of a image based on edge drawing algorithm and generates polylines. These triangle lines could be base lines the fractal grows on.
CycleGAN+Pix2pix
Train the model with >50k pictures of a facade and edges automatically generated from edge drawing. And then draw a facade by hand and see what the AI would provide us. Outputs could be modified by cycleGAN to apply a certain style on. Different style of architecture can be achieved by cycleGAN and experienced with AR technique. Based on the output, we can design the facade afterward. Facade could be any architectural elements like interior design.
Eg.
Changing the mood or style of a space is already achieved in video games using CycleGAN, we could figure out a way to implement it in rendering interface. This allows architects to either decide the color, style of architecture in early design stage or present to client later. This could also help to check if the architecture fits the surrounding context.
TEAM APPROACH - COMBINED PROCESSES
PROCESS 1:
Intent: To combine and coerce two opposing programs of varying requirements within the confines of a tight space using Artificial Intelligence to solve the spatial organisation. Project intents to develop a collaborative design approach between Human and the AI.
Step 1: Curve Path Optimization using Galapagos.
Step 1a: Programmatically define the spatial bubbles of the program within a confined space. Use Lunchbox ML to determine programmatic adjacencies based on necessity or possible symbiotic relationships (eg: C02 produced by automobile factory is used by vertical gardens).
Lunchbox ML can be used to sort complex data based on the preset conditions.
Step 1b: Using Galapagos to optimize the shortest path through the spatial bubble programs.
Step2: Place the spatial bubble program within an input cube/ set of cubes. Run a code to adjust the formation of the cube based on this input optimised curve.
Step2a: Result for 1 input (optimised) curve
Step2b: Result for multiple input (optimised) curves.
Final animation (As done on python) :
Step3: Use complex geometry developed on Maya to inhabit the scaled warped Fractal Boxes.
- Box b)Geometry on Python c)Input Geometry from Maya
Potential Animation of Maya input Model:
PROCESS 2:
Intent: Surface Decoration and Manipulation
Extrapolating manipulated 2D images into 3D volumetric spaces to affect mood and atmosphere
- Set up database of facade images (50k+)
- Extract critical edge of the facade
- Edge Drawing
- Poly Mesh
- Train AI Pix2Pix to recognize facade using facade image and edge lines
- Pix2Pix
- Generate facade by hand drawn facade lines
- Pix2Pix
- Change the color, style and mood of the generated facade
- CycleGan
- Capricious architecture: using AR to suit space for every person
- Zapworks
- Sight Plus
OUTLOOK
Questions
How would we speculate on the AI in design ?
How do we design using our capabilities in AI ?
Exterior Form Massing, Building Plan Generation
Optimizing schematic arrangement to program parameters
- Rethinking real estate: Walls of Change Redux
- Frenemies: symbiotic architecture of conflicting spaces
Comments
Post a Comment