Exploring AI-Fabrication in Shaping the Future of DIY-AT Design: Insights from Makers
Leila Aflatoony, Mixuan Li, Yiyun Zhang, Irene Jacob, Shujian Xu, Ziqi Tang, Andre Grossberg · 2025 · ASSETS 2025: 27th International ACM SIGACCESS Conference on Computers and Accessibility · doi:10.1145/3663547.3746343
Summary
This paper investigates how generative AI can support Do-It-Yourself assistive technology (DIY-AT) design and fabrication. The researchers developed a web-based GenAI interface that combines text-to-image generation (via MidJourney) and image-to-3D model conversion (via Meshy AI) to create a streamlined pipeline from concept to printable 3D model. The interface features a curated keyword selection system organized around activities of daily living (ADLs), product types, adaptation methods, materials, and image styles, designed to help users craft effective prompts without requiring AI expertise. The team evaluated this tool through remote design workshops with 18 DIY-AT makers, including 11 clinical makers (primarily occupational therapists) and 7 technical makers (engineers, designers, and self-fabricators). Participants created hypothetical end-user profiles and used the GenAI interface to design assistive devices for individuals with hand impairments, focusing on tools to support daily activities like eating, writing, cooking, and personal hygiene. Data was collected through think-aloud protocols and semi-structured interviews, then analyzed using collaborative qualitative coding. The study focused specifically on hand impairments because of the critical role hands play in virtually all daily tasks and the high demand for personalized grip aids, utensil adaptations, and other hand-related AT devices. The resulting designs included adapted spoons, toothbrushes, writing grips, bottle holders, cup handles, can openers, and adaptive switches, which were subsequently 3D printed and refined.
Key findings
GenAI demonstrated clear value in sparking creativity during early-stage ideation, with 9 of 18 makers noting its potential as a creative catalyst that helped bridge gaps between imagination and visualization. Clinical makers particularly valued the tool for enabling real-time collaboration with clients, allowing patients to see and respond to visual AT concepts during consultations, which improved engagement and trust. The tool successfully lowered technical barriers for makers without CAD experience, enabling them to generate 3D-printable models without specialized software skills. However, significant challenges emerged: AI outputs tended toward generic, one-size-fits-all designs that failed to reflect the nuanced, individualized needs central to AT design. Makers identified a critical gap in disability-specific training data, noting that AI models produced mainstream consumer products rather than specialized assistive devices. The tool could not handle precise dimensional specifications needed for ergonomic fit, generate discrete or movable parts, or ensure printability in terms of orientation and structural integrity. Clinical makers flagged vocabulary mismatches where professional AT terminology like "u-cuff" was not recognized by the AI. Experience level shaped tool perception: novice makers found it empowering while experienced CAD users found it slower and less precise than direct modeling. Trust in AI-generated outputs remained a consistent concern, especially for clinical makers responsible for device safety.
Relevance
This study provides important insights for the intersection of AI, accessibility, and maker culture. It demonstrates both the promise and current limitations of using generative AI to democratize assistive technology creation. For accessibility practitioners, the findings highlight that AI tools designed for general-purpose use carry inherent biases toward mainstream products, making them inadequate for the specialized, personalized nature of AT design without significant adaptation. The three design directions proposed—addressing data bias through disability-centered training data, expanding accessibility through multimodal inputs and inclusive interfaces, and reframing GenAI as a collaborative design partner rather than an automated solution—offer a practical roadmap for developing more inclusive AI fabrication tools. The paper also underscores that clinical expertise remains essential in AT design; AI should augment rather than replace the judgment of therapists and makers who understand the safety, functional, and emotional dimensions of assistive devices.
Tags: assistive technology · generative AI · digital fabrication · 3D printing · DIY assistive technology · maker culture · occupational therapy · personalization