AI-Driven Storybook Generation for Personalized Autism Support

restricted_to_institution
Restricted (Penn State Only)
Author:
Shah, Riya
Area of Honors:
Human-Centered Design and Development
Degree:
Bachelor of Science
Document Type:
Thesis
Thesis Supervisors:
Syed M. Billah, Thesis Supervisor
Mohamed Khaled Almekkawy, Thesis Honors Advisor
Keywords:
Artificial Intelligence
Generative AI
Autism Spectrum Disorder (ASD)
Natural Language Processing
Human-Centered Design
Abstract:
Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder characterized by differences in communication, social interaction, and behaviors. Story-based interventions have been used widely by caregivers in helping children cope with daily activities and social situations. However, current narrative generation tools for this purpose remain highly inflexible, involve considerable manual work, and lack opportunities for meaningful personalization. This thesis utilizes a research through design framework to study the role of generative AI in helping caregivers create personalized story-based interventions. Requirements identified through interviews with caregivers include personalization support, ease of use, and control over the generation process. Guided by these requirements, a prototype system is designed and implemented that generates structured, illustrated storybooks from high-level caregiver input. The system enables iterative customization of both narrative and visual content to better align with a child’s needs. Through the development of this system, key challenges in translating human-centered design requirements into generative AI systems are uncovered, including issues of visual consistency, sustainability, and cost. These findings highlight both the potential and limitations of generative AI in supporting accessible, personalized autism interventions.