Beyond Shape: Construction-Aware Languages for Design for the Physical World
Date
relationships.isAuthorOf
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Design-for-fabrication systems often ingest conventional geometric representations, like meshes. However, these designs must actually be made in the physical world. How the real objects are made is an integral part of design, as it controls performance metrics like fabrication time, material usage, material properties, and fabrication robustness, among other things. Though geometric representations capture shape and behaviour, they cannot capture these construction details; users are left either doing onerous low-level reasoning and brittle editing, or relying on mythical "sufficiently smart compilers" to generate the fabrication plan they already had in mind. This thesis argues that design systems should instead let users specify construction explicitly, and that doing so requires a language-based design representation built from three components: geometry as a partial input, abstractions of domain-specific fabrication operations, and support for a user-in-the-loop workflow in which the representation is directly editable and results can be previewed end to end. I develop this argument across three systems. The first, for machine knitting, develops abstractions for the fabrication procedure of knit illusions, and creates a system for users to design their own illusions, including novel double-sided illusions. This project establishes the basic principles in a scoped setting where the set of fabrication plans is limited. The second extends knitting to 3D layouts and shaping through a scheduling language over compositions of 2D panels, where a schedule specifies how each panel is composed, decomposed, and knitted. The third carries the same ideas beyond knitting, presenting a domain-specific language for animating content on electromechanical displays, where schedule-like motion effects are applied to keyframe inputs. Together, these systems show that using a design representation where construction is a first-class, language-based component enables the design and creation of real objects where using geometry alone would not.
Description
Thesis (Ph.D.)--University of Washington, 2026
