Our new report includes the generative algorithms which explore the parametric space guided by constraints, objectives and diversity measures. The algorithms can generate designs from scratch, or adapt existing designs using artificial evolution, constrained optimization and quality-diversity search. The intended interaction is aligned with AEC requirements and accounts for cognitive requirements of VR-aided environments detailed as described in previous research for the project.
The currently implemented algorithms and benchmarks provide a large flexibility for the up-coming step of integrating the algorithms in an interactive setting, in the VR-space of the PrismArch platform. Furthermore, the modular architecture of the developed framework facilitates its future extension and the generation and exploitation of Designer Modeling methods that can enhance its operation.
Various workshops and discussions took place for the needs of the research and their outcomes could be summarized in the following shared vision: The design-assistance system, which is based on Quality Diversity (QD) algorithms should be formed as a system that helps the designers to explore the design-space more efficiently, by providing them with design examples that satisfy the specified design constraints, while being diverse along a number of dimensions that are selected by the designer. At the same time, the system’s operation should be guided by the designer’s intentions, preferences and overall judgement. This way, the proposed use of Quality Diversity forms a dynamic cooperation between the designer and the AI, resulting in a novel form of design experience.
Furthermore, the data-traces of the designers’ activity will be used for Designer Modeling: the implementation of machine learning and statistical models for the purpose of identifying patterns in the needs and goals of different types of designers, and also for learning the preferences of individual designers. These models will be used to adjust, filter, and discern when to show AI-generated suggestions to designers in order to maximize their impact to the task at hand.
You can download and read the full report here, under Deliverable 2.2 – Integration-ready version of AI algorithms to traverse the parametric solution space.