Draft of Abstract for ACADIA 2008

multi-scalar reciprocity as a model for bio-inspired design

How can the feedback loops found in natural systems be harnessed to help us navigate the difficult and sometimes conflicting needs of sustainability, economy and delight in architecture?

Biology has been an inspiration to the architecture and engineering professions for centuries because of its elegant resolution of complex problems. As such, the imitation of natural systems – either in look or in operation – is nothing new. Only lately, with greater understanding and the use of computers, have designers been able to mimic more complex natural phenomena for use in high performance designs.

One of the digital techniques being developed lately is the use of localized adaptation to generate form. The enormous potential of this technique, namely the ability of design elements to self-organize, is only partially tapped by most methods. The most common method, based on Darwin’s theory of natural selection, simulates the phenotypic response of an organism to its immediate environment. However it is only a one-way relationship: The part is effected by the specifics of its surroundings, but does not affect the whole.
In actual operation, natural systems have been observed to be much more reciprocal. Not only does the part adapt to the whole at a local scale, but the small result of its role in the system has an aggregate effect: The whole also adapts, sometimes catastrophically, based on the action of many parts. The feedback loop this creates is responsible for the robust self-organization found in many natural systems, and results in the elegant solutions that are so inspiring to architects and engineers.
The process of architectural design already involves feedback loops, as it attempts to negotiate the often-conflicting requirements of sustainability, economy and delight. Traditional iterative methods are effective, but a computational process abstracted from the study of natural systems would be extremely valuable for use in complex high-performance designs. Such digitally assisted feedback loops would allow designers to more easily find design solutions which satisfy extremely ambitious sustainability goals and resolve them with subjective design criteria.
This research builds off of the theoretical groundwork of Georges Canguilhem, in his discussion of the “Milieu”, and of Darwin’s and Lamarck’s theories of biological morphogenesis. It also draws from Reyner Banham’s writing on the use of mechanically controlled temperature gradients to define space (in place of physical barriers). Steven Wolfram’s Cellular Automata and their distillation of part-to-part relationships down to a set of very simple rules also provides some inspiration in designing an algorithm to achieve the feedback loops described above. Alan Turing’s work on chemical morphogenesis and reaction-diffusion equations is also of some interest.
To develop a computational design methodology for architecture, I have studied biological systems that exhibit the desired feedback at a variety of scales. I found that many such systems rely on chemical cues – for example chemotaxis – to transfer information between separate parts. This ambient communication medium is physiologically independent of the communicating organisms, and has a global aggregate presence as well as a direct local one. As such, reciprocal feedback is facilitated across a multiplicity of scales. The chemical cue system is easily abstracted to an algorithmic model in the form of data packets – small pieces of information combined with location that are both created and consumed in a dynamic milieu.
The focus of this research is the development of a digital design methodology, based on the above study of chemical communication and feedback, which employs a milieu of ambient information to achieve multi-scalar reciprocity. The research includes a study of relationships between elements found in selected natural systems, and assessment of which relationships are analogous to those found in architectural projects. The result is a computational methodology, using scripting, to simulate the localized awareness of parts in complex natural systems. Proof of concept is shown as the digital methodology is applied to the Solar Decathlon – a national competition to design a small residential project with high performative requirements.

The next step in applying the cellular model to a residential project: The Solar Decathlon house with its many performative requirements. This exercise explores the capacity of a single module to adapt to changing needs at a local scale, where it may be a window, door, roof exterior decking, or interior finish.

Basically unmodified, the braced frame generated by the script is used to support a prefabricated panelization system that takes on different qualities according to its location in the building: solar panels on top, windows on the sides, little feet on the bottom.

Very Basic, however, I have come to the point where I begin to see the strategy clearly. My experience at SG2008 and the resulting algorithm were good inroads into the use of functions, and global variables to handle different tasks, all called by a central script (called “Solid” in the diagram below).

The next step is to reformulate the above strategy for a process that is more specifically aimed at the ideas I am exploring in my Thesis, namely using bottom-up process to manage architectural qualities, in a recursive feedback loop.
The beginnings of such a recursive strategy are shown in the diagram below: an older version that I developed in the beginning of the semester based on Cellular Automata. It will take some more in-depth work to combine all of the previous work into a single method but all the pieces are here. they Just need to be assembled (what am I talking about with all this recursive stuff? please see my latest Abstract, in column at left).

The BMW Welt in Munich by Coop Himmelb(l)au

The 1972 Olympic Stadium by Frei Otto (right across the street)

Smart Geometry is an annual conference and workshop (this year in Munich) that explores the cutting edge of parametric and computational design. It’s mainly about Bently’s parametric plug-in for microstation: Generative Components (GC).

Just like Christmas: 3D prints of projects from the past 3 days’ workshop arrived today

Andrew Kudless’ Voronoi model above

Now some Pics of some of the stuf I’ve been playing with in GC:

This is all based on the neighborhood script that identifies points adjacent to the one in question and gives them an identifier number (listed in diagram below). The surroundings can then be analyzed and the point populated with geometry that is responsive to local context. In this case, the response comes in the form of very speciic connections, but its just a test case for a technique that can be used to aggregate geometry in a way that’s responsive on many levels – dealing with issues of structure, energy efficiency, and perceptual environment.

Some components to play with techniques of connection :

Close-up showing how the pieces fit together

Below, the “Symbolic Model” of my scripting Strategy. “Solid 01″ is the component being aggregated. The various graph variables and functions decide where and how it happens.

That same component aggregated on points along two adjoining surfaces – notice how it connects them.

And in a double tower

Tomorrow, I’m off to BMW World!

Transcription Factors, Genes and Protein Production

The following is paraphrased from An Introduction to Systems Biology by Uri Alon, Taylor & Francis, London, 2007

Cells live in a complex environment and can sense many different signals including physical parameters, such as temperature and barometric pressure, biological signalling, molecules from other cells, beneficial nutrients and harmful chemicals. Cells respond to these signals by producing the appropriate proteins to act on the internal and external environment.

Transcription factors

Transcription factors are special proteins designed to change rapidly between active and inactive molecular states . When active, they bind directly to DNA to regulate where, and how it is read (transcribed). Genes (DNA) are transcribed into mRNA which is then translated into proteins which can act on the environment. The new proteins act on the environment by forming new tissue, or sending biological signals which are picked up by other transcription factors.

from An Introduction to Systems Biology

Since they are tuned to respond to pre-set combinations of environmental signals, Transcription Factors are essentially symbolic representations of the various environmental states that the cell might find itself in. Evolution has selected internal representations that symbolize states that are most important for cell survival and growth. These transcription factors regulate their their target genes to mobilize the appropriate protein response to each combination of symbolic states.

Transcription Networks

Transcription networks are formed of transcription factors acting on their target genes as well as looped reactions: transcription factors acting on each other. There are two types of loops: feed-back and feed-forward (see diagram below). Feed back loops happen when the proteins produced by the gene (Z below) act on transcription factors (Y below), which in turn act on other transcription factors (X below), which act on the initial gene. Feed-Forward Loops happen when a transcription factor (X below) acts on both other transcription factors (Y) and 1 or more genes (Z). In the network diagram below, a representation of roughly 1/3 of the transcription network happening inside e-coli bacteria, arrows represent lines of effect.

from An Introduction to Systems Biology

Transcription factors and Turing Patterns

Transcription factors can act on genes as either activators or inhibitors, increasing production of a given protein or decreasing it. Here, they share a commonality with Alan Turing’s theory of mathematical biology and morphogenesis, which depend on the conflicting action of chemical activators and inhibitors in a reaction-diffusion system. All this means that biological systems form patterns based on chemicals that both cause and repress change. The patterns depend on how fast those chemicals diffuse through the system, and how fast they degrade. Examples of simple Turing patterns include the spots on a leopard or the computer generated patterns below, it is theorized that tuned properly, Turing patterns could yeild any shape or form.

2D and 3D Turing Patterns

from Computational Studies of Pattern Formation in Turing Systems, by Teemu Leppänen, a publication of the Helsinki University of Technology Laboratory of Computational Engineering

Relevance to This Study

The diagram below shows a simplified distillation of the feed back and feed-forward loops in cell transcription networks. Transccription factors recognize combinations of pre-set environmental states and release the appropriate chemical signals, as well as influencing genes to make new proteins. Chemical signals as well as new proteins act on yet other transcription factors and the loop continues.

If we replace the environmental factors above (specific to living cells) with the “environmental” factors that could be specific to parts of a building, then we get a similar diagram (below). In this case “environmental” simply means qualities of other parts found in the immediate vicinity.

In the above diagram, “transcription factor” has also been replaced by “case”: scripting jargon for a symbolic representation of predetermined conditional factors. “Gene” has been replaced by “generative script”: the part of the code that creates new geometry in the same way that genes create new proteins. “Signal molecules” have been replaced by “signal particles” or points with data attached to them that can influence the case.

My hope is that a script that is based on the above diagram can combine the bottom-up design process exhibited in the cellular automata script (see previous post) with a new top-down process for more direct control over some aspects of the design.

Today we had a pinup review with Andrew and 5 other Faculty. Our task, for studio, was to develop a schematic massing for a cellular colony of buildings – about the size and shape of the solar decathlon houses. The units were to be off-grid and able to be delivered to the site by truck. The idea was that we’d take one of the ‘cells’ and develop it into the solar decathlon project (There’s more about the program on the class blog).

The research I’d done into Cellular Automata scripting served me well in developing an aggregation strategy for the units. I simply devised a new set of 6 rules, loosely based on typical CA operation, but specific to architecture.

Each rule was given a probability, meaning it would only happen a percentage of the time. This gave the pattern a degree of randomness and dynamism where, otherwise, the buildings would form a very regular grid almost immediately as the southern exposure rule took precedence. Below is an example of the script working:

The reviewers’ response to my project was basically “cool process, but its implementation, ie: the ideas behind the actual project, need some work”

The best point of the day for me was when I really “got” the point behind this phase of the studio: working by looking for the synergies between seemingly conflicting programmatic goals and sites.

In short, the process can be distilled into these 5 verbs:

Analyze, Distill, Extract, Adapt, and Apply

My work in the next few weeks will focus on other self organizing systems (like the inner workings of biological systems and urban development patterns) and see what phenomena can be brought over and turned into a script.

More soon…..

Cyclic Cellular Automata examples

R1/T3/C3/NM

R2/T4/C5/NM/GH

The notation of Cyclic Cellular Automata has the “R/T/C/N” form, where:

R – specifies the neighbourhood range (1..10).

T – specifies the threshold – minimal count of cells in the neighbourhood having the next state, necessary for the cell to change to that state.

C – specifies the number of states in the rule (0..C-1).

N – specifies the neighbourhood type: NM stands for extended Moore, NN for extended von Neumann.

GH – Greenberg-Hastings Model: A prescribed number of colors N are arranged cyclically in a “color wheel.” Each color can only advance to the next, the last cycling to 0. Every update, cells change from color 0 (resting) to 1 (excited) if they have at least Threshold 1′s in their neighbor set.

In general Cyclic CA rules should be started from uniformly randomized boards.

From Cellular Automat Rules Lexicon
(http://psoup.math.wisc.edu/mcell/index.html)

Possible Scripting Strategy:
These rules can be adapted to a cellular aggregation of small residential units by equating the states (C) to types of construction rather than color. one state (the 0 or black state above) will represent no construction, and serve as space for light, air and access between buildings.

I thought the following rules could also be added:
Roughly half of the units aggregated will be double height.
No units will be built directly to the North of a double height unit.

Here’s a start, more soon:

Download my first foray into C.A. rhinoscipting here.

Click here to see an animation of the pattern evolution.

This is based on the simplest of C.A. – Just a 2D pattern in which each tile updates itself incrementally based on an assessment of its nearest neighbors- a popularity contest. There are four types of tile with varying degrees of open-ness. Each tile matches itself to the type of tile found most prevalent among the 8 surrounding tiles.

It starts with a randomized field…

…..and ends with a simple grouped pattern.

At the moment, this amounts to nothing but negative entropy (order from disorder) but it’s got potential. Just imagine the possibilities when the tiles are building parts and the rules get a little more complex. hmmmmm..