HCII Ph.D. Thesis Proposal: Humphrey Yang

Computer-aided design (CAD) tools have become widespread and integrated into our social-technical infrastructure in the past few decades. From revolutionizing the manufacturing industry in the mid-twentieth century to the recent celebration of generative artificial intelligence, they have expanded and augmented our imagination and ability to materialize ideas. Their usefulness has made them integral to our modern design and engineering workflows to tackle complex and contextual problems.

Active materials arise as a new episode of CAD tool development. These materials are dynamic: they often have spatiotemporal behaviors activated by external stimuli. These materials are abundant in nature, leveraged for survival strategies, and have started to manifest in the manufactured world. They exemplify an “embodied material intelligence” that differs from that resulting from neurons, consciousness, and cognition. In HCI, these materials have been leveraged to augment, program, and enrich interactions between humans, computing agents, and the living environment. Noticeably, their design requires spatiotemporal thinking and modeling, prediction, and prescriptions of dynamic behaviors, making it particularly challenging to both experienced and novice users. Thus, we often develop CAD tools with specific abstractions and workflows for active material systems. However, these CAD tools often have rigidly defined functions and assumptions and navigate a confined design landscape, lacking the flexibility to tackle real-world, contextualized design problems. The tools also fall short of capturing and supporting the ambiguity and dynamism of design objectives inherent to iterative design processes. Consequently, it is difficult to use them to create designs that are satisfactory at both quantitative and qualitative values, making it challenging to use these tools to respond to real-world challenges.

In this thesis, we argue that the challenges could be overcomed by fostering human-CAD tool collaboration in active material design. This sentiment calls for a new modality of human CAD-tool interaction that de-emphasizes computers as automation. Instead, we highlight their potential as collaborative partners that help us make informed decisions in an iterative design process and as a “compass” to navigate design spaces. We believe that this modality of CAD tool could assist designers in creating active material designs that are not only numerically performative but also qualitatively satisfactory in other values like aesthetics. We use compliant mechanisms design - an emerging mechanical design paradigm synergizing with active materials - as a context to manifest and explore this idea while pushing the technical boundary of active materials design. Through the CAD tools we built, we demonstrate that tools built according to the sentiment could help designers create designs that respond to real-world scenarios. The subsequent chapter proposes exploring reinforcement learning as a CAD toolmaking paradigm for navigating highly complex and nonlinear design spaces, using interconnected compliant mechanisms as an example.