an exploration towards a more material-informed architecture. New developments in reinforcement learning become the starting point for a piece of software that supports non-linear modes of design, constantly oscillating between human and machine intelligence. Targeted specifically at timber construction, the work aims to develop a design methodology as well as an architectural formalism, which allows for the reuse of non-standardized timber parts.

in the first iteration of the software, an algorithm was created with the freedom to tweak an input design on a local scale within a given threshold.  each input design is structurally analysed. this density gradient gets translated into a series of lines, representing the length, orientation and thickness demanded for potential building elements. the evolutionary optimisation agen t  iterates through thousands  of options to optimize the best match for each line in correspondance with the material catalog the main issue with this approach was that pieces were aggegrated without a cohesive connection logic, making the physical assembly logic unviable

the global organisation of the structure is set through a spacefilling tesselation. The tetrahedral structure can reconfigure through recursive subdivsion. An advantage of this approach is that while the scale and density of the structure may vary depending on structural needs, the angles between individual members will always be contained within a limited set of options.

in order to minimize the amount of robotic manipulation necessary on-site, the connection logic for the main load bearing structure was ultimately chosen to be steel joint able to be used in up to 30 different scenarios. the connection is designed to transfer loads as directly as possible from one timber element to another. future iterations are planned to reduce the complexity in its geometry.

the approach deals not only with the functional viability of urban mining, but aims to explore culturally what it means to create new architecture with the help of demolishing cherished buildings of the past. in playing with this contradiction, the architecture becomes an archive of the past in the literal and cultural sense. an architecture that aims to visualize the traces of its materials through a formal reinterpretations / defamiliarization of the buildings it originates from. first steps in this approach used deep learning algorithms to create new artificial facades, whose aesthetic is derived from common visual patterns shared by all previously destroyed buildings.

the stylegan2_ada neural network was used in a further step to guide the original assembly algorithm, creating new facades driven by the formal reinterpretation of the cities lost heritage and the constructional limitations reusing existing materials.

the algorithm is tested on a site in vienna, used to automate the planning of a construction that exclusively uses a limited set of existing materials. the methodology can be interacted with by a designer through geometric input and programmatic demands. the final structure is a more materially-aware design, created through the collaboration of artificial intelligence and human intuiton.