Emergent Design Martin Hemberg Imperial College 2004 Emergent
Emergent Design Martin Hemberg Imperial College 2004 Emergent Design
Who's that guy and what's he doing up there? ● ● ● Martin Hemberg Developed Genr 8 with the Emergent Design Group (EDG) Work at Emergent Design + Technologies (Em. Tech) Architectural Association (AA) Martin Hemberg Imperial College 2004 Emergent Design
Agenda ● ● Evolutionary Computation Artificial Life Emergent Design Motivation Evolutionary Computation (EC) ● Artificial Life (Alife) ● EC and Alife combine well for design tasks Example applications – Design Martin Hemberg Imperial College 2004 Emergent Design
Motivation, Architecture ● Break new grounds in architecture – ● ● ED uses a different logic Natural form has aesthetic and functional values Hard to obtain using ordinary methods Martin Hemberg Imperial College 2004 Emergent Design
Motivation, Computer Science ● ● Application of EC – Exploration, not optimisation – Fitness evaluation Use computers creatively – Beyond CAD-tools – Requires new algorithms Martin Hemberg Imperial College 2004 Emergent Design
Evolutionary Computation ● Optimisation method – ● Inspired by natural evolution – ● Randomized Population adaption generation by generation Comes in many flavours: GA, GP, ES, GE, etc Martin Hemberg Imperial College 2004 Emergent Design
Neo Darwinian Evolution ● Survival of the fittest ● Selection on phenotype – Through environment ● Genotypic inheritance ● Reproduction ● Blind variation Martin Hemberg Imperial College 2004 Emergent Design
Artificial Evolution Pseudocode for an EA: generation = 0; initialize population; while generation < max-generation evaluate fitness of population members for i from 1 to population-size select two parents; Fitness biased selection crossover parents -> child; Inheritance mutate child; Variation insert child into next generation’s population; endfor; generation++; update current population endwhile; Martin Hemberg Imperial College 2004 Iterate by generation Emergent Design
Selection ● Selection ensures that fitter individuals have a higher probability of being selected for the next generation – Tournament – Proportional Martin Hemberg Imperial College 2004 Emergent Design
Fitness ● ● ● A leap from natural evolution A quantified numerical value is assigned to each member Try each member on the problem and rank them or quantify their performance Martin Hemberg Imperial College 2004 Emergent Design
Fitness Evaluation ● ● How to assign fitness according to aesthetic criteria? – EA are good at finding optimal solutions – Need to figure out what to optimize Open problem Martin Hemberg Imperial College 2004 Emergent Design
Fitness Evaluation, strategies ● Rule based – ● Learn user preference with neural network – ● Too many parameters, fails in practice User acts as fitness function – ● Hard to define and encode rules Human fatigue, short runs Co-evolve critics Martin Hemberg Imperial College 2004 Emergent Design
Fitness Evaluation, my view ● Put the user in the loop ● Create tools with the designer in mind ● – Make them open-ended – Can't predict user's need and context Parameterized fitness function – User has high level control of evaluation – Fitness emerges as a combination of factors Martin Hemberg Imperial College 2004 Emergent Design
What is Artificial Life • How does life arise from the non-living? • What are the potentials and limits of living systems? • How is life related to mind, machines, and culture? Martin Hemberg Imperial College 2004 Emergent Design
Two definitions of emergence ● ● ● The whole is greater than the sum of the parts Emergence is the phenomenon wherein complex, interesting high-level function is produced as a result of combining simple low-level mechanisms in simple ways. Examples include brain, society Martin Hemberg Imperial College 2004 Emergent Design
Alife and EC for Design ● Evolutionary computation – – ● Creative and generative qualities Discovery and adaptation more than optimization ALife – – Agents interacting with environment can model elements of design and conditions of the problem Emergent properties in outcome from bottom-up approach The ‘fit’ is powerful and good, but not perfect Martin Hemberg Imperial College 2004 Emergent Design
Surface Component System ● Simple growth model ● Select tiles from a predefined set – Rules for which tiles are allowed Incorporates structural analysis in the EA – Martin Hemberg Imperial College 2004 Emergent Design
Using the tool ● ● Implemented as a MEL script FEA in Ansys Martin Hemberg Imperial College 2004 Emergent Design
Geometric Fitness criteria ● Fast and easy to evaluate and understand Number of support points – Support point distance – Height – Holes – Martin Hemberg Imperial College 2004 Emergent Design
Structural Fitness criteria ● FEA is computationally costly – Don't evaluate each generation Martin Hemberg Imperial College 2004 Emergent Design
Genr 8 – A Design Tool for Surface Generation ● ● Combines EC and an organic growth model Surface are grown in a reactive simulated physical environment Martin Hemberg Imperial College 2004 Emergent Design
Lindenmayer Systems ● ● ● Organic growth model Widely applied to model plant growth in computer graphics L-systems are important in formal language theory Martin Hemberg Imperial College 2004 Emergent Design
Rewriting systems ● ● A set of production rules are repeatedly applied to a seed Rules are expressed as a grammar Seed: Rule: a a->ab b->ba Martin Hemberg Imperial College 2004 Emergent Design
Turtle Graphics ● Turtle graphics is a way to visualize the grammar – Rules are interpreted as instructions for moving and drawing in 3 D space Seed: Rule: Angle: Martin Hemberg a a->a+a--a+a 60 Imperial College 2004 Emergent Design
![Plant Models ● Operators “[“ (push state on stack) and “]” (pop state from](http://slidetodoc.com/presentation_image/47c4c1a7b33f18c0c8d1448621adf353/image-25.jpg "Plant Models ● Operators “[“ (push state on stack) and “]” (pop state from")
Plant Models ● Operators “[“ (push state on stack) and “]” (pop state from stack) allows branching – Time delay – Stochasticity – Environmen (tropism) Seed: Rule: Angle: Martin Hemberg a a->a[+a]a 45 Imperial College 2004 Emergent Design
![Map L-Systems a -> d[~a]b a c b a b seed b b c](http://slidetodoc.com/presentation_image/47c4c1a7b33f18c0c8d1448621adf353/image-26.jpg "Map L-Systems a -> d[~a]b a c b a b seed b b c")
Map L-Systems a -> d[~a]b a c b a b seed b b c -> b[-~a]b c b b d Martin Hemberg a b c d b a b -> b b b a b d d -> c b b b d b b b a b d b a c b b c b d b a b b c a b b d b b b c b b b B b c Imperial College 2004 b Emergent Design
HEMLS ● 3 D ● Scaling ● More complex productions Context sensitivity – Time variation – Stochastic – Martin Hemberg Imperial College 2004 Emergent Design
Environment ● Forces ● Attractors – Repellors – Gravity Boundary – Martin Hemberg Imperial College 2004 Emergent Design
Evolution ● Search the universe of possible surfaces – Find a grammar corresponding to the surface that the designer has in mind – Explore the universe to find interesting forms Martin Hemberg Imperial College 2004 Emergent Design
Grammatical Evolution ● Automatic generation of grammars – Very hard to construct by hand ● Many constraints -> problematic for GP ● Grammatical Evolution allows any language – Use Backus-Naur Form (BNF) to map linear genome into a grammar – Genetic operations are separated from language Martin Hemberg Imperial College 2004 Emergent Design
Mappings ● Genr 8 contains several mappings – ● ● Increases the complexity Individuals represented by linear genome Selection on the phenotype that is expressed through an environment Martin Hemberg Imperial College 2004 Emergent Design
Design Evaluation and Fitness ● Fitness function with multiple parameters – – – Size Smoothness Soft boundary Subdivisions Symmetry Undulation Martin Hemberg Imperial College 2004 Emergent Design
Fitness Criteria ● User determines target values and weight for the criteria – Multiparameter optimization – Trade-off between criteria – Population gives a family of solutions Martin Hemberg Imperial College 2004 Emergent Design
Interruption, Intervention and Resumption (IIR) ● ● Traditionally, EA are monolithic User can guide the evolution by interacting and interfering – – Allows for greater control The tool cooperates with the user Martin Hemberg Imperial College 2004 Emergent Design
Using Genr 8 ● ● Set up environment Define fitness criteria and other parameters Run a few generations Analyze the results, adjust parameters and environment Martin Hemberg Imperial College 2004 Emergent Design
More on Emergent Design ● Genr 8 website http: //www. ai. mit. edu/projects/emergent. Design/genr 8/ ● EDG website http: //web. mit. edu/arch/edg/ ● Em. Tech website ● http: //www. aaschool. ac. uk/et Martin Hemberg Imperial College 2004 Emergent Design
- Slides: 36
