Image 01


beyond the Kuiper Belt, over the sea

Serializing Architecture

January 30th, 2009 by Kit

As usual, BLDGBLOG blows my mind.  Skim over all the word-clouds—those are old-hat by now—and take a look at the last two paragraphs of the post.  The idea of making room clouds is great, but what it sparks in my mind is even more interesting, to me: the idea of serializing architecture.  If you could find a good encoding for architecture, you could use all sorts of existing techniques to play with large amounts of architectural data. I am thinking, particularly, of using genetic algorithms to grow building designs that are highly fit for various purposes; that would require a good fitness algorithm, too, but seriously, the field of computational architecture is untapped!

So how would one serialize a building?  I suspect that the work of Karl Sims is not irrelevant; serializing a building is a lot like serializing a creature.  The only difference for use in a genetic algorithm is the fitness function, which for a creature is based on reproductive ability before total energy loss, but for a building… well, that’s the question, isn’t it?  Clearly it would have a number of factors: safety, efficiency for purpose, pleasantness.

What would you think was important to a building’s fitness?

Tags: , ,

  • I think the first factor you’d have to model is the use of the building, so you need a fitness function that has people moving between task nodes within the building. Time wasted pathing between task nodes might then be included as bad for fitness (at least if you believe architects like Susanka and her “not so big” idea for home design). Energy efficiency as you mention might be a good factor too, although that’s largely taken care of by building materials over layout (with big caveats like needing to know the weather of the building site, window size, etc.).

    I think it’s computationally feasible, but you’d have to stack a number of models on top of each other to evaluate the fitness of a given serialized building, which might make it that much harder to understand what is actually being optimized.

  • kit

    I had actually meant the “task efficiency” that you describe, not energy efficiency. But energy efficiency is another consideration! Room for rooftop gardens, anyone?

    I think your point is a good one, though: there are enough dimensions to a building’s fitness that it could be hard to tell what areas are getting optimized within your (presumably) hand-weighted values for the dimensions.

  • kit

    also, do you know you have a mountain dew bottle stuck through your head?

  • I suppose the short answer is trial and error – develop several competing models for task efficiency (presumably based off of real-world data) and optimize all of them. The fitness model that allows more population members to converge on high fitness is to some extent then validated. Throwing more computation at the problem might be considered cheating, however :oD.

  • I am aware. Which wordpress default are you using to retrieve that?

  • kit

    It is just grabbing your gravatar. I forget where I set that up.

  • Ben

    This is hauntingly reminiscent of some of the possible applications of the research I’m just getting started with now. Basically, we’re working on a way to make user-driven evolutionary algorithms (like these “biomorphs” applets) less stupid, and what you describe is a possible application — if, of course, you have a representation language for architecture.

    I wonder, however, if the “room clouds” comment was intended on this level, or more on the level of a basic inventory of room types.

  • kit

    I’m sure the BLDGBLOG post was intended as an inventory; I was elaborating on it. But that sounds like interesting work.

  • Ben

    It occurs to me to add that the result would be a graphical interface to a machine learning algorithm that tries to design the house you want by playing Plenty Questions.