Exploring AI-Assisted CAD for Occupational Therapists
Written by Zekun Chang
What happens when the assistive device someone needs does not exist?
Occupational therapists often support people whose needs are highly specific to their abilities, routines, environments, and goals. Commercial assistive devices can be expensive, and in many cases, the right product may simply not exist. When that happens, therapists may need to modify the activity goal or continue relying on caregiver assistance. 3D printing offers a potentially lower cost and more flexible way to create customized devices. However, designing printable objects usually requires CAD and fabrication skills that many occupational therapists do not have.
This summer, as a Siegel PiTech PhD Impact Fellow working with YAI, I explored whether AI-assisted 3D design tools could help bridge this gap. I focused on AdamCAD, a tool that allows users to create and modify CAD models through natural language.
Studying how Occupational Therapists design
I conducted an exploratory formative study with three occupational therapists and one person-centered technology specialist who participated in a pilot session. The occupational therapists had between two and more than 30 years of clinical experience, but none had previous CAD experience.
Participants were asked to think of an assistive device that could help someone they had supported and to use CADAM, a commercially available text-to-3D CAD tool, to develop a 3D concept while thinking aloud.
Their designs included a toothpaste dispenser for someone with limited fine motor control, an angled utensil adapter to support independent eating, a hand-worn zipper aid for someone with limited grip strength, and an enlarged adaptive grip for a standard utensil.
Figure 1: Participant-created assistive device concepts in AdamCAD. From left to right: a wall-mounted toothpaste dispenser, an angled utensil adapter, an adaptive utensil grip, and a hand-worn zipper aid.
Text-to-3D lowered the barrier, but communicating design intent was still challenging
All participants found the text-to-3D interaction relatively easy and straightforward. Despite having no CAD experience, the occupational therapists could quickly begin turning ideas into visible 3D models without first learning conventional CAD commands.
The challenges became more apparent as participants tried to refine their designs. They often had a clear understanding of the functional goal, but struggled to describe it in a way the software could interpret accurately. Participants sometimes lacked the technical terms needed to describe mechanisms or geometric features and searched Google or used other AI tools to find relevant terminology.
For example, one participant wanted to create a grip that fit more naturally around a person's hand. She described the desired shape as a "ball," and the AI interpreted the request literally, generating a spherical form. Another participant revised her prompts several times while trying to communicate a particular angle and orientation.
These examples suggest that text-to-3D can make modeling more accessible, but users may still need support expressing functional needs and spatial ideas in ways the system can accurately translate into geometry.
Visualization-supported Design Thinking
The immediate 3D visualization was also valuable. For one participant, talking through the problem with AI and seeing the model helped transform an initially vague idea into a more concrete solution. Instead of needing to fully imagine the final device before beginning, she could react to what appeared on screen and refine the idea. In some cases, participants also felt that the AI suggested solutions they had not initially considered. This suggests that AI-assisted CAD may be useful not only for producing models, but also for supporting design exploration.
From a 3D Model to a Fabricatable Device
Another challenge emerged around fabrication. The occupational therapists primarily focused on whether a design addressed the person's functional need. Exact dimensions, materials, robustness, and printability were not always considered early in the process. One participant initially included the fork itself in the design before realizing that the available 3D printing material might not be appropriate for an eating utensil. The concept then shifted toward creating only an adapter that could connect to a standard fork. The pilot technology specialist provided an interesting contrast. Because she had more familiarity with 3D printing, she asked questions about size, printability, materials, and structural robustness, and modified her design based on the limitations of the available printer. This highlighted two complementary forms of expertise. Occupational therapists understand people, activities, accessibility needs, and functional goals. Fabrication requires additional knowledge about geometry, materials, mechanisms, and manufacturing.
Zekun Chang
Ph.D. Student, Information Science, Cornell University
Looking Ahead
The study suggests several opportunities for future AI-assisted design tools.
Systems could ask more guided questions about the person, activity, dimensions, existing objects, materials, accessibility, and fabrication constraints. Participants also wanted more contextual visualization, such as seeing a hand or existing object alongside the generated design. AI could also help users find language for ergonomic forms and mechanisms rather than expecting them to already know technical design terminology.
Future work could continue exploring how AI can better translate person-centered needs into geometry.
A second direction is CAM (Computer-Aided Manufacturing) and fabrication. Creating a digital model is only one part of the process. Future tools could also help users determine whether a design is printable, what materials are appropriate, and whether the available fabrication process can actually produce it.