Inverse Discrete Design

Abstract

Historically, engineering and manufacturing have been dominated by analog processes—manual, continuous, and heavily reliant on human intuition and trial-and-error. Digital technologies have revolutionized communication and computing, enabling unprecedented scalability, precision, and complexity, yet our methods for designing and fabricating physical structures remain largely analog, fragmented across disciplines, and resistant to integration, significantly limiting innovation across complex, multidisciplinary domains. We need a parallel digital revolution in the physical realm that implements principles of discretization, modularity, hierarchy, and error correction to create new possibilities for physical design and fabrication. By applying these principles to physical structures, we can enable precise placement of functional materials with embedded electrical and mechanical properties. These `Digital Material' structures bring digital programmability to the physical realm, mirroring the transformative impact of digital technologies in other fields.

This transformation requires new design methodologies to faithfully model our complex environment and enable us to design a new world where the physical and digital become indistinguishable. My thesis introduces a fully declarative and inverse workflow for designing and building scalable Digital Material systems. These workflows allow users to model and design systems that span scales (micro, meso, macro) and disciplines (electrical, mechanical, aerospace, architectural engineering) without being experts in all—or any—of these fields. The introduced workflow uses domain knowledge as priors and universal design representations across all stages of design, simulation, optimization, fabrication, and control.

Leveraging the discrete and hierarchical nature of the Digital Material system, I first introduce a multi-scale multi-physics simulation tool for mechanical metamaterial. I then present target-based multi-scale optimization workflows for the design of the geometry as well as the growth of cellular structures and soft robots. Next, I introduce advanced path planning algorithms—swarm, recursive, and hierarchical—for scalable robotic swarm assembly; as well shape-control optimization for the reconfiguration of these robots. The work extends to introduce a comprehensive design tool to design and build intelligence and physical computing, and culminates in an end-to-end inverse design workflow of electromechanical structures, that transforms text specifications into physical designs. The workflow is used to design a plethora of static and dynamic structures, ranging from bridges and shelters to aerospace structures, robots, and electronics.