The aim is to create significant variations. Designs that are very different to each other. Every body knows how to do parametric models now, we trying something else here. Trying to get computers to help create :
1) Designs that are very different to each other
2) Designs that are viable
As you know it is easy to create 1000's of junk designs. We are attempting to build good "schema" for generative design. The quality of generation depends very much on the scheme. That is the touch part. So far there is not much out there on how to build "expressive models". Models that are capable of representing a very diverse set of designs. let me know if you come a cross any material on this. I will be very interested.
Well then I guess my question would be, how do you define how different a design should be and how do you define viable in your system? Are these decisions in the hands of the user or in the scheme you discuss?
In this method, the decisions are in the hands of the user. We are allowing the user to control the levels of "creativity" which traslates into a percentage of variation. Random dimensions are generated within these limits and used to create the variations. The way the initial dimensions are set will determine the nature of the variations and this is different in different modles and is entirley up to the designer.