Summary: Researchers adapted a children’s exercise program using AI-generated, simplified instructions to better serve neurodivergent learners. By applying ChatGPT to more than 500 activities from the InPACT video series, the team clarified and standardized movement directions to make the program more accessible and inclusive.
This adaptation aims to expand equitable access to brief, structured physical activity—especially for children with autism spectrum disorders and other neurodivergent conditions—by combining universal design principles with generative AI. The project continues to grow with plans for additional resources and translated materials to reach a wider audience.
Key Facts:
- AI adaptation: ChatGPT assisted in rewriting over 500 exercise instructions to be clearer, more concise, and more inclusive for neurodivergent children.
- InPACT at Home: Originally developed during the COVID-19 pandemic to support at-home learning, the program was updated with consistent, easy-to-follow cues to support universal design.
- Sensory and motor support: Instructions were adjusted to accommodate balance, motor coordination, body awareness and language-processing differences.
Source: University of Michigan
A University of Michigan research team used AI to improve home-based exercise “snacks” for kids with autism and related conditions.
Led by exercise physiologist Rebecca Hasson from the U-M School of Kinesiology, the research team revised an existing brief-activity program that had been adapted for home use during the pandemic. Their goal was to make the program more accessible for neurodivergent children by simplifying language and standardizing cues across activities.
The team used ChatGPT to revise instructions for the program’s 132 exercise videos, extracting and refining directions for more than 500 discrete activities. Their findings appear in Frontiers in Physiology.
Hasson began developing ways to integrate movement into children’s daily routines in 2012 as part of efforts launched during the “Let’s Move!” campaign. The resulting InPACT (Interrupting Prolonged Sitting with Activity) program promotes short, three- to four-minute “exercise snacks” that can be embedded into the school day; it has been implemented in 25 Michigan schools.
When the pandemic forced learning into homes, the research team adapted InPACT for broadcast and digital delivery. Working with the Michigan Learning Channel and the state Department of Education, they aired the program on television to reach families without reliable internet. Within six months the show drew an estimated 15,000–20,000 daily viewers.
Assistant professor Haylie Miller, a developmental psychologist and movement scientist, observed that the original video instructions were tailored toward neurotypical learning styles and could be difficult for neurodivergent children to follow. Neurodivergent learners often process sensory input and language differently and may need clearer, more explicit prompts to succeed with movement tasks.
To address these differences, the team set out to adapt the InPACT content using a universal design approach. Undergraduate researcher Tania Sapre led the initial instruction revisions and explored whether generative AI could accelerate and standardize the process.
Sapre described experimenting with ChatGPT to shape and format directions. She realized the tool could help turn complex movement descriptions into succinct, repeatable prompts that teachers, caregivers and other researchers could use or adapt for new exercises.
The researchers first organized the video content and identified more than 500 activities across 132 videos. They grouped activities into skill-based categories—jumping, core, lateral, sport-oriented, upper body, lower body and compound movements—to structure their queries to ChatGPT.
Using targeted prompts, they asked the AI for simplified, step-by-step instructions and for condensed cue words suitable for neurodivergent children—for example, requesting both a full breakdown and a shortened set of steps for a jumping jack. The team then reviewed and refined each AI-generated set of instructions to ensure safety, clarity and alignment with InPACT’s principles.
A central design rule was the “Three C’s”: consistency, conciseness and clarity. The team emphasized simple, elementary sight words, consistent phrasing across movements, and breaking complex actions into smaller, manageable components to reduce ambiguity and cognitive load.
Alanna Price, a regional health coordinator who helped revise the directions, stressed the importance of adaptive physical education: modifying activities so all children can participate helps build motor skills, strength and social-emotional well-being. Adaptations often include simplified cues, visual supports and sensory-friendly activity options.
The researchers are also developing a starter pack of activity play cards for learners who require more foundational practice before progressing to the full InPACT routines. They aim to be proactive in creating accessible resources and to welcome feedback that improves inclusion.
Future plans include translating the adapted videos and materials into Spanish and Arabic—the two most commonly spoken languages in Michigan after English—to broaden reach and equity.
Study co-authors include Alanna Price (Detroit Public School Community District); Anna Schwartz and Kerry Winkelseth (U-M School of Kinesiology); Ron Zernicke (U-M School of Kinesiology and U-M Medical School); and Leah Ketcheson and Jeanne Barcelona (Wayne State University).
About this AI, ASD, and exercise research news
Author: Morgan Sherburne
Source: University of Michigan
Contact: Morgan Sherburne – University of Michigan
Image: The image is credited to Neuroscience News
Original Research: Open access. “Enhancing home-based physical activity for neurodivergent children: adapting the InPACT at Home program with AI and universal design” by Rebecca E. Hasson et al., Frontiers in Physiology
Abstract
Enhancing home-based physical activity for neurodivergent children: adapting the InPACT at Home program with AI and universal design
Purpose: Many community- and home-based physical activity programs are not explicitly designed to meet the support needs of neurodivergent children, contributing to reduced activity levels. This study aimed to adapt the InPACT at Home program so neurodivergent children can safely and confidently participate in short, structured physical activity at home.
Methods: The team used a rapid-cycle adaptation process that included: (1) organizing and categorizing video content by skill and movement type (problem exploration); (2) assembling an expert group to guide instruction development (knowledge exploration); and (3) employing generative AI to draft concise instructions and cue words for each activity (solution development). Experts then refined the output using the Universal Design for Learning principle of Representation to ensure multiple ways of understanding the movements.
Results: From 132 InPACT at Home videos, researchers identified more than 500 activities and grouped them into core categories: jumping, core, lateral, sport, upper body, lower body and compound movements. Meetings with experts reinforced the need for consistent, concise and clear language. AI prompts such as “Provide simplified step-by-step instructions for a jumping jack, suitable for a neurodivergent child” and “Condense the step-by-step instructions for a jumping jack, suitable for a neurodivergent child” were used to generate and refine instructions.
Discussion: By combining AI-driven text generation with universal design principles and implementation science frameworks, the adapted program seeks to increase equitable access to brief, structured physical activity for neurodivergent youth. Next steps include evaluating whether these adaptations boost participation in InPACT at Home and increase overall physical activity levels among neurodivergent children.