Collaborative Research: HCC: Small: End-User Guided Search and Optimization for Accessible Product Customization and Design

The widespread availability of inexpensive 3D printers, knitting machines, laser cutters, embroidery machines, and other desktop manufacturing devices has the potential to drive a revolution in the creation of bespoke, person-specific manufacturing. However, designing artifacts for manufacturing currently requires a great deal of expertise and the ability to use advanced computer aided design tools.

This project seeks to put the power of design in the hands of end users using generative artificial intelligence and guided search algorithms. This project's framework will enable domain experts, and even end users, to create domain-specific design tools without any programming expertise. For example, a clinician might use the project's framework to make a tool for generating custom hand-splints that fit the needs of specific patients with hand injuries.

The research team will focus on the important domain of accessibility devices, physical objects that address the varied accessibility needs for people with disabilities and related health concerns. The project's work will culminate with a design workshop with stakeholders from the disability community who use mobility devices (seated and standing). The benefits of this project include potentially enabling many people of different abilities to find the equivalent of bespoke solutions without the costs of custom design.

The ability for domain experts to rapidly create generative design tools that address custom and specific product needs for end users would be transformative for accessibility, product development, and the consumer economy more broadly.

To put optimization-based search in the hands of end users, the project will enable domain experts without programming expertise to specify three critical components of generative design using the project's novel framework, Metamorph. The research team will emphasize interpretability throughout the work. First, domain experts need to be able to describe the design space over which the generative design tool they are creating should search.

While Metamorph is domain agnostic, the project's evaluations will focus on digital fabrication domains. Thus, the project will support a range of design representations including computer aided design and learned embeddings created using deep learning tools. The research team will develop novel approaches for visualizing the design space to support verification by domain experts.

Second, domain experts need to be able to describe the metrics of the search in terms of both the design space and any end user input. As part of this effort, the research team will develop the first database of parameterizable optimization metrics for domain experts, and an interface for letting the user select, parameterize, and compose them, or teach by example.

By treating metric specification as an end-user programming problem, the research team can apply new techniques to metric specification, such as learning and composition, and will emphasize interpretability in presenting the results back to domain experts. Finally, the project will support domain experts and end users in selecting appropriate optimization algorithms validating and verifying the resulting domain-specific design tool that Metamorph produces.

This task will involve first using qualitative methods to understand the best approaches to interpretability in the optimization space, followed by iterative design to implement and test approaches informed by this work.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.