Join us on Tuesday 19 December for the PhD proposal presentation of Pedro Veloso.
SoA PhD PROPOSAL PRESENTATION
TITLE: designLOOP: an agent-based system for architectural composition
By: Pedro Veloso, PhD-CD Candidate
Dr. Ramesh Krishnamurti, SoA CMU, Chair
Dr. Gianni Di Caro, SCS CMU Qatar
Dr. Reinhard König, Computational Architecture, Bauhaus-University Weimar
This research addresses the design of spatial composition with online agent-based models.
In conventional design processes, designers use feedback and problem restructuring to solve a spatial composition, which results in a task environment that is dynamic and unreliable. Besides, the large number of explicit and implicit constraints in a compositional problem makes it too complex to solve by deterministic algorithms. In contrast, multiagent models have a potential to explore good solutions precisely in these dynamic and unreliable task environments.
After describing the general approach to design problems in automated space planning (ASP), I review the precedent and recent works (2008-2017) in multiagent space planning (MASP), exposing current trends and gaps in the field. Then, I present a theoretical framework and an implementation guideline for a MASP research construct called DLOOP. This construct integrates two complementary agent-based design systems: an arrangement of discrete spatial units and a circulation network. The spatial units are represented by predominantly convex polygons structured as a spring-physics entity. These units can interact and navigate in the compositional space to satisfy local constraints such as adjacency, access, alignment, and visibility. The circulation network is a graph connecting the spatial units with direct and indirect circulation elements. It is generated from a complete set of edges and nodes defined by the boundaries of the spatial units.
In the end, I organize and describe the next steps of this research, including technical specifications, the tests to validate the research hypothesis and a timetable with all the future tasks.