The theme for Smart Geometry 2013 London is Constructing for Uncertainty and I am a participant in the Digital Intuition and Prediction cluster championed by Mirco Becker, Stylianos Dristas, David Kosdruy & Juan Subercaseaux. The Smart Geometry conference is focused on a 4-day workshop where 100 international participants come together to develop ideas and techniques that are at the forefront of digital design. This cluster is particularly interesting for me as it relates to my PhD research. The agenda of the cluster lies in the belief that we can augment intuition by enabling prediction via computation. A framework for achieving this considers a design problem as a series of interrelated modules: design generation, analysis, evaluation and optimization.
“Starting from a design context of a complexity that is either undrawable or unsolvable by parametrics, novel and specific descriptive systems must be found. Secondly, the design will be evaluated by building a custom analysis system or linking to existing frameworks / solvers. Thirdly, in the optimization phase we explore heuristics, AI, and simulation methods to balance equilibrium between desires and affordances. By this time we will have built a system that describes, evaluates and optimises the design context. Lastly, in the systems-design phase we interrogate this set-up by testing it for sensitivity and robustness, and conclude on its ability to augment intuition.”
GENERATIVE SPATIAL PERFORMANCE DESIGN SYSTEM
My “unsolvable” task is related to architectural spatial design, which can be considered a wicked problem and can have a multitude of solutions for any given design brief. This is heightened at the conceptual stage of design, where a large number of solutions are created and need to be evaluated in a short period of time. This notion of a ‘solution driven’ design process is a common approach to solving design problems and provides expertise to resolve the complex interrelated design variables. Whilst this can be achieved, there is an argument that the repertoire of organisational patterns, design precedent knowledge and the precise criteria and computation of spatial evaluation required for generative exploration is more than what can be expected from the accumulated knowledge of an experienced architect. This project seeks to augment the designers intuition through the construction of a design support system focused on spatial performance. This Generative Spatial Performance Design System is described below:
Given a sketch design with an initial spatial configuration, the intent of the system is to have the capability to search a corpus of design precedents for comparable designs, and utilise the precedent knowledge to generate informed variations for conceptual spatial design exploration. The system can be decomposed into a series of inter-related modules: spatial knowledge capture, design filtering, generative design exploration and evaluation.
SPATIAL KNOWLEDGE CAPTURE
A key component of the system is the structuring of precedent / sketch designs into a specific format suitable for spatial analysis. This is achieved through the construction of a Parametric Spatial Configuration Rig composed of a series of spaces and a network graph identifying the building configuration.
Once the Parametric Spatial Configuration Rig has been constructed, the building can be spatially analyzed. A series of metric are calculated including standard room measurements as well as spatial configuration analytics derived from Space Syntax techniques. These analytics can be visualised on the precedent / sketch design for immediate spatial configuration feedback.
To make the system effective, a corpus of design precedents need to be captured and stored into a custom designed spatial database. This database can be interrogated for comparable designs which in itself is a multi-objective search criteria that can be tailored by a designer’s preference.
Once a design precedent has been determined as a suitable target, the system should be able to generate a series of alternate informed spatial configurations. Currently the system can achieve this through two approaches. The first compares a designers initial sketch to the target precedent found on the spatial database. The two buildings are functionally mapped together to allow the formation of a hybrid network which can be biased towards the source or target design. The image below shows an equally weighted hybrid configuration.
Whilst the first approach attempts to form a new novel configuration, the second approach assumes the target precedent to be the desired configuration and uses this as the source for generating new designs. The process for generating new designs involves a parametric rig that reconfigures the network into a new building footprint. This reconfiguration seperates out the spatial components into the new domain, however connects them with “springs” to ensure they have the same configuration topology. An optimisation routine is then applied that dynamically relaxes the new layout into novel design configurations that matches the target configuration in a new context.