Chia Y. HanECECS Department

University of Cincinnati

Kai LiaoCollege of DAAP

University of Cincinnati

Collective PavilionsA Generative Architectural Modelling for

Traditional Chinese Pagoda

CAAD futures 2005

Introduction

• Complex Adaptive Systems (CAS)• Computational approaches based on CAS

• Inspired by ‘bio-logic’• Categorized into non-classic, connectionists and natural

computation

• Applications for shape computation, architecture and urban morphology• formalization of natural form through fractal geometry• modelling of animal behaviour patterns, e.g., the Boids

algorithm• biological growth processes, e.g., the L-systems• evolution & adaptation phenomena, evolutionary

computation, etc.

Background

• Current avant-garde architectural practice• the works of Greg Lynn, ETHZ, etc.

A Sequence of Similar Modules with Iteration and Interaction

Background (cont.)

• Design Computation research• Recursive algorithms for shape

computation, e.g., shape grammar • Fractals in architecture and urban

structure • Evolutionary design for architecture • Artificial life for architectural design and

2D/3D Cellular Automata for building plan and mass/volume composition

Background (cont.)

• Problems with current CAS approaches• Unclear association to the architectural form &

space concepts, and architectural space theories,

• Lack of an in-depth, systematic analysis of design manners that provides a holistic and connectionist view,

• Insufficient development of aesthetic theory and historical perspective of the new paradigm.

Background (cont.)

• Can current CAS approaches do this?• Iteration: Shape grammar? Structure?

Needs and Proposed Solutions

• To upgrade the concepts of architectural form and space based on ‘bio-logic’, self-organization and non-linear order –

• To develop a framework of generative architectural modelling that is applicable to design analysis & criticism, and formal & spatial design.

• To study how the shape patterns/components are used as the basic entities for architectural design/modelling –

• To integrate basic shape with architectural space concept and spatial patterns in architectural settings.

• To discover how past architectural works can enrich future design using generative methodologies in architectural modelling –

• To study traditional Chinese architectural structures, in particular, pagoda, to help us gain new aesthetic knowledge and develop historical research methods for enriching design manners & architectural vocabulary for current design practices.

Western vs. Eastern philosophy

Evolution of Chinese Pagoda

Embedded into terrain

Pagoda Forest

Array Formation

Our Approach

1. Study both the architectural form-making and space design analysis based on CAS viewpoints

2. Incorporate generative design and evolutionary computation in implementation

3. Provide both global and local considerations through multi-agent modeling and simulation

Our Method

• Two level abstraction (space and form) for descriptive & generative model, combining:

- graph-based space description with - recursive shape computation

c

Basic pavilion parts

RoofBracketWall/columnPodium/banister

Variants of composition

Generative Model – Profile characteristics

Sample global spatial structures

1-5-5-1 1-5

Pavilion units

4-sided 6 8 round

Variation by Eave Feature

Adjacent pavilions interacting as agents with living behaviors

Moving

Producing

Extending eave

Degenerating (roof)

Sample composition patterns (Rhythm/ emergent social structure)

Algorithm

Algorithm

PhaseI. – Generate pavilions 1) Initialize design space for pavilion units 2) Randomly select one of the distinctive formal

patterns of pavilion units and record its topographical graph

3) Randomly seed a set of pavilion units genotype (initial pavilion population)

4) Decoding the genotypes into phenotypes for evaluation of the fitness (subjective)

5) Put the genes of selected pavilions into the pool of evolved genes for the production of pavilion population

PhaseII. Generate spatial patterns 6) Initialize design space for the spatial patterns of

pagodas / the topologies of pavilion layouts. This includes coding of the topologies of pavilion layouts, initialization of parameters, and creation of working environment

7) Randomly seed a set of graphs for the spatial patterns of pagodas / the topologies of pavilion assembling (initial pavilions topology population)

8) Decoding the genotypes into phenotypes for evaluation of the fitness (subjective)

9) Put the genes of selected pavilion layouts into the pool of evolved genes for the production of pavilion assembling population

PhaseIII. Generate assemblies of pavilions 10) Initialize design space for the assemblages of

pavilions as the predecessors of design populations: selecting one of the genes of selected pavilions, and a set of the genes of selected pavilion layouts, and creation of working environment

11) Randomly specify one of the genes of selected pavilions and a set of the genes of selected pavilion layouts, so as seed a set of assemblages of pavilions as the predecessors of design populations, by recursive computation using the selected pavilion as axiom for graph-based grammar design

PhaseIV. Generate pagodas 12) Initialize design space; coding of the

interactions between pavilions, parts of pavilion, coding of global and local constrains, initialization of geometrical parameters; and creation of working environment

13) Consider a pagoda as a collection of living pavilions: let individual pavilion move, grow, shrink, produce, die, and interact with others to generate a set of pagodas as design populations

PhaseV. Search pagoda design space 14) Specify the graph-based design grammars

(genotypes) of desirable pagoda pairs for evolution from the viewpoint of subjective, aesthetic standards, and creation of working environment

15) Evolve genotypes: Crossover, mutate and reproduce. A set of evolved genotypes and their correspondent phenotypes are the generated pagodas.

Start

Phase II

Phase I

Phase III Phase V EndPhase IV

Flow Chart

Phase I – Generate pavilions (local features)Phase II – Generate spatial patterns (global features)Phase III – Generate pavilion assemblies (integration)Phase IV – Generate pagoda (adaptive refinement)Phase V – Select final configuration (explore design space)

Phase I - Generate Pavilions (GA)

1. Initialize design space for pavilion units2. Select a formal unit pattern and record its

topological graph3. Seed a set of pavilion units genotype

(initial population)4. Decode genotypes into phenotypes for

subjective evaluation of the fitness, accept or continue

5. Add the genes of selected pavilions into the pool, evolve, and go to step 4.

Phase II – Generate spatial structure

1. Initialize design space – topologies of pavilion layouts

2. Randomly seed a set of graphs for topologies of pavilion assemblies

3. Decode the genotypes into phenotypes for subjective evaluation, accept or continue

4. Put the genes of selected pavilion layouts into the pool, evolve, and go to step 3

Phase III – Generate assemblies

1. Initialize design space for combining phases I & II (forming a collection of pavilion genes and layout genes)

2. Generate an assembly from the above pool, using a selected pavilion unit as axiom and applying recursively operational rules according to the selected layout

Phase IV – Generate pagoda

1. Consider a pagoda as a collection of living pavilions, explicitly encode the following: interactions between pavilions, local and global constraints, and geometric and form parameters of the pavilions.

2. Do local refinement by letting individual pavilion move, grow, shrink, produce, die, and interact with others to generate a candidate version of the pagoda

Phase V – Exploring design space

1. Specify aesthetic standards for selection

2. Invoke phase IV to generate a set of pagoda candidates, and select a pair with desirable characteristics.

3. Evaluate them subjectively, and let them evolve further in the design space to generate newer versions

Exploring design space

Compositional rules for plan layout

recursive

Shanghai Jin Mao TowerBy Skidmore, Owings & Merrill

Genes for Shanghai Jin Mao Tower

∑

Sample result - glPagoda

Conclusions

• Investigated generative architectural modeling for Pagoda and traditional Chinese architecture

• Explored and extended the potential of adaptive computing for architectural design methods

Contributions

1. Provide a study of both the architectural form-making and space design analysis from the CAS viewpoints

2. Incorporate generative design and evolutionary computation in implementation

3. Provide both global and local considerations through multi-agent modeling and simulation

Thank you !

Temple of Heaven

Twin Pagodas

Pagoda Triplet