Talk to a Wall explores the intersubjectivities of human and machine in architectural design methods.
A neural network was trained (LSTM) by procedurally sketching architectural ideas. The network learnt from the human generated drawings to then search in a high- dimensional space and understand the associations between the parameters and the resultant drawings. The model was then able to draw in a similar style to that of the human when triggered by an initial set of parameters. It will also propose its own variation based on the learning data set. The model predictions would then be translated into robotic motion parameters to drive the kinematics of a 6-axis industrial robotic arm with the aim of creating a truly collaborative human – machine drawing.
A Collaboration with:
Vishu Bhooshan – ZHACODE
Cristobal Valenzuela – RunwayML
Jose Luis Garcia del Castillo y Lopez – Autodesk