🤖 AI

Accessibility Scout: Personalized Accessibility Scans of Built Environments

This paper presents Accessibility Scout, an LLM-based system that generates personalized, scalable accessibility assessments of built environments from photos by tailoring evaluations to individual users' specific mobility needs and preferences.

Original authors: William Huang, Xia Su, Jon E. Froehlich, Yang Zhang

Published 2026-01-23

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: William Huang, Xia Su, Jon E. Froehlich, Yang Zhang

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are planning a trip to a new city. You want to visit a cool café or stay in a nice Airbnb, but you use a wheelchair, a walker, or get tired easily. Right now, finding out if a place is actually accessible for your specific body is like trying to guess the contents of a sealed box by shaking it. You might see a photo online, but you can't tell if the door is too narrow, the bed is too high, or if the noise level will give you a headache.

Current tools are like generic "checklists" that ask, "Is there a ramp?" But they don't ask, "Is the ramp too steep for your specific wheelchair?" or "Is the bathroom too small for your walker?"

Enter "Accessibility Scout."

Think of Accessibility Scout as a personalized detective that uses a super-smart AI brain (called a Large Language Model) to look at photos of buildings and tell you exactly what you will face when you walk (or roll) through the door.

Here is how it works, broken down into simple steps:

1. The "Body Blueprint" (User Modeling)

First, you teach the AI about yourself. You don't have to fill out a boring medical form. You can just chat with it or type things like, "I use crutches for short trips but a wheelchair for long ones," or "I get tired easily and need a handrail," or "I can't hear well in noisy rooms."

The AI turns your words into a digital blueprint of your body and needs. It's like giving the detective a custom map of your specific challenges.

2. The "Mission Briefing" (Task Identification)

Next, you show the AI a photo of a place (like a restaurant or hotel) and tell it what you plan to do there. Maybe you say, "I want to go on a date here."

The AI breaks this down into tiny steps, just like you would in real life:

"Okay, to have a date, you need to walk to the table."
"Then you need to sit down."
"Then you need to reach for your drink."
"Then you need to talk to your partner."

3. The "Red Flag" Scan (Accessibility Scans)

Now, the AI looks at the photo through the lens of your Body Blueprint and your Mission. It acts like a spotlight, highlighting potential trouble spots.

Generic AI: Might just say, "There is a table."
Accessibility Scout: Says, "Warning! The table is too high for someone in a wheelchair to reach their drink, and the chair is fixed, so you can't pull it out to sit down."

It also spots things other tools miss, like "The lighting is too dim for someone with poor eyesight" or "The TV in the bedroom will be too distracting for someone who needs quiet."

4. The "Practice Run" (Collaborative Learning)

The system isn't perfect yet, so it asks for your help. It shows you the red flags it found and asks, "Is this right?"

If you say, "Yes, that's a problem," the AI learns.
If you say, "No, I can actually reach that," the AI updates its blueprint.

It's like a video game where you train the AI. The more you play with it, the smarter it gets at understanding your specific needs.

What Did They Actually Find?

The researchers tested this system with real people who have different mobility needs (like using wheelchairs, walkers, or crutches) and looked at 500 photos of real places.

It works: The system successfully found problems that generic checklists missed. For example, it noticed that a high bed might be fine for a tall person but impossible for someone transferring from a wheelchair.
It's personal: When the AI used a person's specific "Body Blueprint," the warnings it gave were much more useful than generic warnings. People felt the personalized warnings were actually helpful, while the generic ones felt like useless noise.
It's fast and cheap: The system can scan a photo in about 10 seconds and costs very little to run, meaning it could eventually scan thousands of places on Google Maps or Airbnb instantly.

The Bottom Line

Accessibility Scout is a tool that turns a static photo of a building into a personalized preview of your experience. It doesn't just tell you if a place is "accessible" in a general sense; it tells you if it's accessible for you, based on how you move, what you need, and what you care about.

The paper concludes that this approach helps people with disabilities feel more confident exploring new places, reduces the stress of the unknown, and gives them a voice to tell AI exactly what they need. It's not a magic wand that fixes the building, but it is a powerful flashlight that helps you see the obstacles before you even leave your house.