Worlds Within Worlds Co-evolving designers and critics Bio-inspired design processes Gregory Hornby, UC Santa Cruz Jordan Pollack, Brandeis University.

Wo

rld

s W

ith

in W

orl

ds

Co-evolving designers and critics Bio-inspired design processes

Gregory Hornby, UC Santa CruzJordan Pollack, Brandeis University

Hod Lipson, Cornell University

Conclusions

• Evolutionary design– Open ended, creative– Challenged by scalability, vague design goals

• Evolve designers, not designs– Generative systems that capture design rules

• Evolve critics that represent user tastes– Provoke users to learn their goals & preferences

• Algebra of user models and designers– Use multiple models to influence multiple designers

Evolution

Lipson & Pollack, Nature 406, 2000

X-band antenna for NASA's ST-5 Mission

Evolving Photonic Structures

With Preble, Gondarenko, Robinson, Physical Review Letters, May 2006

Kinematic Synthesis

Peaucelier (1873)Silverster-Kempe (1877)

• Evolutionary design– Open ended, creative– Challenged by scalability, vague design goals

Evolving Designers

Encoding designers with Modularity, Regularity and Hierarchy

Design Program: Executed Instructions:

Graphical version:

Evolving Table DesignersEvolving tables: fitness = height*surface area*stability/material.

No MRH enabled: MRH enabled:

Evolved TablesTable fitness = height*surface*volume / material

Families of Designs

Height: 8 16 24 32

• A single design program can be used to evolve a family of designs:

Evolving Critics

A Simple Critic

Confidence vs. Uncertainty

Walter Benjamin

Louisville, KY, USAFabLab, Pretoria, South Africa

Rockefeller Univ., New York, USAScience Museum, London, UK

Watch band and Lego™ tire printed on a Fab@home

User 1U

ser

2

User 1

Algebra of user models

Conclusions

• Evolutionary design– Open ended, creative– Challenged by scalability, vague design goals

• Evolve designers, not designs– Generative systems that capture design rules

• Evolve critics that represent user tastes– Provoke users to learn their goals & preferences

• Algebra of user models and designers– Use multiple models to influence multiple designers