Wednesday, November 30, 2005

Genetically Designed Stuff

Friends recently pointed DYWSC? to a start-up firm out of Singapore that claims to be designing barstools, coffee makers, electronics, and watches with something that sounds like genetic algorithms. They appear entirely technically involved, and admirably are designing the software to push their ideas. If this isn’t a prank, it sounds super-cool, and like a great business model for the youngsters… except “Genometrically” isn’t a word
1. The client provides a basic CAD model or pictures and a description of desirable design outcomes.
2. The product will then be modeled Genometrically. We will generate some sample design variations for the client to confirm the desirability of the design outcomes.

3. The model will be further refined using this information. We would then generate any number of variations, which would be presented online as interactive 3d models or as rendered images for the client's selection.

4. The CAD model for the selected design will then be given in the requested format to the client.
5. This is offered as a web based service.

3 Comments:

Blogger Norman Blogster said...

Another interesting blog entry!
This all started in the '80's when William Latham teamed up with IBM to make evolutionary art. Google it for refs. He hand-picked images to breed with each other, so his fitness function was purely aesthetic rather than functional.
Genetic algorithms (GAs) work when you're trying to search extremely large solution spaces, are willing to accept sub-optimal results in reasonable time and want to be surprised. Their success lies in a) defining a reasonable fitness function - i.e. how an individual in the population is judged better to survive than another, so it has more chance of breeding, b) the coding of the genome (similar to the DNA in humans), and c) the mapping from genome to phenome (the end result). The problem with having a fitness function that is purely aesthetic is that the designer has to pick from many alternatives at each iteration in the evolution. So (s)he can't ever really come up with something original.
In other words, you know beforehand roughly what you're aiming for, so why not just design it like that in the first place? I saw a lecture once on someone who used a GA to design a table. Guess what? It looked like a table! I was underwhelmed in the extreme.
However, if you start using functional rather than aesthetic fitness functions, along with intelligent genome to phenome mapping, then you can be genuinely surprised with some way-out designs. I guess this is the real logical ultimate conclusion of the modernist maxim "form follows function". John Frazer did an interesting unit at the AA about 12 years ago based on this and wrote about it in his "An Evolutionary Architecture".

Are you still there?
Ok, back to the point: The best thing about this company is their name methinks, because it's all styling and no substance. If a designer just wants to click on several individuals on the screen to breed, then it's just sad. If they want to write the mappings, fitness functions and coding to try and make something original, then that's laudible and extremely difficult, but may well develop something genuinely original and I look forward to the day that this happens, if I'm still breathing.

Finally, they claim the Blibao museum was modelled in SolidWorks and generated using Gennovate, but I thought Gehry's workshop used Pro Engineer or Catia?

8:19 AM  
Blogger J said...

DYWSC? has been struggling for two years on how to apply Genetic Algoritms to architecture. It comes with much debate, as contributers to the blog spent time working on this, which models city growth by its zoning envelopes. There was even debate dedicated to when success can be measured.

Such algorithms tend to be perfect for optimization. Optimization is generally the tool of engineers (less expensive, lighter, stronger, higher), and not explicitly architecture. What is it that architects seek to optimize? The Vetruvian Triad -- Comoditas, Utilitas, e Venustas – would have us optimizing beauty along with firmness and economy. That’s tough to select for in GAs.

Of course the famous stopping problem should be referenced here too. Once you get a screen full of designs, which one do you choose? The fellas in the post were very unselfconscious about leaving it to the client. Wierd. Sounds like "I'm just the pencil."

7:41 PM  
Anonymous Jasmeet said...

Good news, Genometri is not a prank! I hope you like the sound of it. There is a list of products that you can now download and try for yourself. Check them at genometri.com

GA works on the principle of evolving a winning design from a plethora of contestants. That is it is intended to produce "the best" design, not to provide you with a choice of designs. Two the variation in the design output from a GA can range from very simplistic to very complex product shapes that may not be manufactured.

There is also an inherent problem that lies in translation of the genome to the phenome. It is very difficult to numerate aesthetically pleasing characteristics, for this reason a human interface would be required to choose good looking designs. That is where you add bias and lose serendipity.

Although Genometri does not work directly based on a GA, it uses the concept of a DNA, where mutations may produce very exciting and differing results. Oh, the Bilbao museum was not generated using genovate but its (rendered) version on the genometri's site was :P.

Jasmeet

8:20 PM  

Post a Comment

<< Home