Since July 2015, a number of us in the new AEC Generative Design group have been looking at ways to extend and complete the promise of Project Akaba's algorithmic design generation, so a couple of months ago we started Project Fractal.
Project Akaba had a lot of internal code to generate space planning options from a building program in different ways, with the original "random domino" algorithm delivering the results most people have seen either in my earlier post or in visuals like this:
What we haven't much talked about much outside Autodesk except in a few private briefings and a couple of public events are some other option generation methods we tried, such as the "seed sower" algorithm that scatters room "seeds" and encourages them to grow to the limit of the growth of adjoining spaces...
...the "stuffer" algorithm that follows a series of strategies for dividing up a known space...
...or the more sophisticated goal-seeking recursive optimizations using simulated annealing to find an acceptable space fit within a perimeter, as in the video below:
The drawback to these experiments is that while of increasing interest, they didn't admit customization or easy inclusion of conditions beyond the initial goals and constraints coded directly into the generating algorithm. Obviously, such an approach is inadequate for AEC. Nearly every building project is a bespoke assembly of commodity and crafted objects to answer unique needs that will probably never combine in exactly the same way again. We had to start thinking about accommodating varying conditions, goals, and constraints in a way that would allow designers to conserve and apply their expertise to generate a whole range of options to investigate and develop.
Next up, what we did last summer.