Learning from Form // Machine Learning


Using neural networks in OWL for Grasshopper, create an algorithm that uses supervised learning to achieve categorisation.

Project Objective:

Learning from Form explores two strategies that give insight to form, teaching the algorithm to categorise them based on certain parameters. The first compares the volume with the surface area of the geometry and the second, teaches the algorithm to understand the difference between planar and curved surfaces. In this case, it has been achieved through the inspiration of bird boxes, but it could be applied to any geometric forms.

Strategy One: Volume and Surface Area

This strategy allows categorisation based on two parameters. Using the visualisations to compare each of the geometries.

  1. Calculate the volume of the geometry and the volume of the bounding box of each geometric form
  2. Divide the volume of the geometry by the bounding box volume
  3. Calculate volume, separated into small, medium and large
  4. Feed the data into the tensor sets and compute the algorithm

Results: To achieve better results or more accurate results the algorithm would need to be train more.


  1. Swallow’s Nest: Pink Surface
  2. Weaved Nest: Blue Surface
  3. Bat Box: Green Surface
  4. Traditional Bird House: Yellow Surface

Surface Area

  1. Small: Yellow Box
  2. Medium: Blue Box
  3. Large: Red Box
Strategy Two: Planar and Curved Surfaces
  1. Categorise the bird boxes based on surface type
  2. Using the component Is Planar, retrieve the true or false results for each surface
  3. In the list, replace False with 1 and True with 2
  4. Add all the values in the list together and divide by the list length
  5. Feed data into the tensor sets and compute the algorithm

Results: This algorithm worked very accurately in deciding on the type of surfaces.

  1. If the value is 0, all surfaces are planar (RED box)
  2. When the value is 1, there is one curved surface (YELLOW box)
  3. If the value is 2, all the surfaces are curved (BLUE box)


Learning from Form is a project of IAAC, Institute for Advanced Architecture of Catalonia developed in the Master in Advanced Architecture 2019/20 by Fiona Demeur and Faculty: Mateusz Zwierzycki