Platform

Collaboratory

How Can Data Aggregation Help Disabled Travelers?

Technical Explainers

Disability & Accessibility

Tesla Wells

2025-01-22

Technical Explainer header image of Tesla Wells and an image of wheelchair

IN A NUTSHELL:

If you are disabled, researching the accessibility of a specific location is likely a big part of your travel routine. To help disabled travelers plan their trips, Mobi.AI uses our data aggregation technology to enrich our location datasets with accessibility information. In this technical explainer, we discuss how we can apply different data aggregator tools to the problem and explain where they are limited. We also explore how working with accessibility data foregrounds specific technical challenges with data verification and communicating data uncertainties. Our intent is that by creating this dataset, we not only improve Mobi’s products, but we provide a foundational and needed resource to disabled communities as well.

When we talk with disabled travellers, one of the greatest hurdles we hear about in planning a trip is finding the accessibility information related to their disability. People spend hours manually compiling information. Maybe you are looking at photos of the hotel entrance from different angles to see if they have a ramp. Maybe you are using a screen reader to see if a museum has media for blind patrons. Or perhaps you are reading air quality reports for a certain region to ascertain if a hike is possible with your asthma during your summer break. And after all that research, there is no guarantee that the trip you are researching will even fit your needs—maybe you learn that an island you were hoping to relax on has no accessible hotels after personally vetting each one. You might have to start all over to find a place that can meet the unique abilities and constraints of your body.

Of course, sometimes this information is not available because it has not been documented. There will, inevitably, always be instances where you must call a restaurant to see if they have a bathroom that can be used by a wheelchair user because that information is not online. But a lot of this data does exist online (and new data is being added everyday!), and if it exists we want to give you access to it. Mobi.AI uses cutting edge data enrichment technologies to sleuth out and centralize this type of information from unconventional sources. We are hoping to make meaningful progress on expanding current disability-centered databases using this technology. This database would allow us to make specific recommendations—like finding users a hotel, restaurant, or activity that fits perfectly in their trip. It will also help with trip discovery—allowing us to identify cities or regions that are highly compatible with a user’s accommodation needs. On a technical level, doing this requires that we first familiarize ourselves with what data to collect, extract this information from data sources, enrich our location database, and then assess and convey how certain we are of our data’s accuracy.

We look at multiple sources to construct a starting point for “relevant information.” First, we look to ADA accommodation and travel guidelines to help us identify different impairment categories and common travel problems stemming from disability. This allows us to map out different communities of people who face constraints when they travel (some non-exhaustive examples being mobility impairment, air-quality sensitivity, and deafness).

Image of Tesla Wells

In disability contexts, it is especially important to provide verification or qualifications for our data’s accuracy/veracity. While sending users to a pizza restaurant instead of a general italian restaurant might be annoying, this type of inaccuracy is, at the end of the day, fairly harmless. In contrast, the consequences of booking a shuttle that is not wheelchair accessible could leave a traveller stranded in a foreign country.

Tesla Wells

Algorithm Engineer

Image of Tesla Wells

In disability contexts, it is especially important to provide verification or qualifications for our data’s accuracy/veracity. While sending users to a pizza restaurant instead of a general italian restaurant might be annoying, this type of inaccuracy is, at the end of the day, fairly harmless. In contrast, the consequences of booking a shuttle that is not wheelchair accessible could leave a traveller stranded in a foreign country.

Tesla Wells

Algorithm Engineer

Once we have made this taxonomy, we go searching for common requests, accommodations, and constraints associated with each community. This means looking at online groups of disabled travelers to see what information is typically crowdsourced, working directly with disabled people interested in varying types of travel, and looking at what information is collected by other companies focused on disabled travel. Our Data Development team researches potential data sources and uses data aggregation and enrichment tools to add disability-related information to our POIs (Points of Interest).

To do this, Mobi uses many of the tools detailed in our “What Makes Mobi.ai’s Data So Special?” technical explainer. In this piece we explain what kinds of “unconventional data sources” Mobi’s data teams are able to process and how we compact it all into one coherent database. We can tell you if the building has a wheelchair ramp from photos on the internet. We can quickly scrape both the website and reviews to tell you how tactile that art exhibit really is. We can crunch how AQI varies within space and time to find a hike that doesn’t leave you winded. Applying these data processing techniques in disability contexts would be fairly novel. While research on the mapping accessibility data is not new, research on the intersection of “data enrichment” and disability is a developing field (one that we’re hoping to contribute to). For example, this research paper on processing geo-spatial data to collect information on urban accessibility only came out in the last year.

In disability contexts, it is especially important to provide verification or qualifications for our data’s accuracy/veracity. While sending users to a pizza restaurant instead of a general italian restaurant might be annoying, this type of inaccuracy is, at the end of the day, fairly harmless. In contrast, the consequences of booking a shuttle that is not wheelchair accessible could leave a traveller stranded in a foreign country. When disabled travellers book pre-planned tours the services being offered are not just finding lodging, food, and activities that are disability-friendly. Part of the appeal of a local guide or tour service is their ability to verify that the information is credible and up to date. Choosing to process information related to disability would require us to hold our data quality to similarly high standards.

What does scaling vetted locations look like when it’s done well? We look at Wheel the World (WtW), a database for wheelchair-friendly travel, as an inspiration. WtW created their database through meticulous, manual, in-house documentation, and verification of accessibility features—a process that is labor intensive, but offers stronger guarantees of information accuracy. We also look at the long history of crowdsourcing disability-mapping work and discussion of their efficacy in both activist and academic contexts. This research establishes both the volume of information required to properly vet accommodations and the standards our data aggregation needs to hold itself too. If we aren’t manually vetting accommodations, we must have processes in our data aggregation/processing to preserve our information sources. A few years ago, this was only possible if we meticulously stored and tracked specific information sources. This was sometimes difficult when machine learning techniques were used to enrich data. However, the release of large language models (LLMs) over the last few years has renewed interest in tracking information sources and including citations in the world of AI.

We hope to leverage both old and new techniques to understand the accuracy of our data. In practice? If you need proof that a hotel has a ramp, we can link you to the photo where we used machine vision to detect a ramp so you can judge yourself. We should also be able to let you know “this information is based on data that is six years old, so it might be out of date.” If we don’t have information, we should also let the user know—and then hopefully provide the user with the hotel’s phone number and email, so that they can call themselves to ask!

Diagram of data ingestion

But as we said in our other data-oriented technical explainers, having killer data is only the beginning because data is the foundation on which all of Mobi.AI’s other tools can operate. When we have good data about places, we can create better itineraries and schedules. If we have data that tells us the museum tour with a signing guide happens on Tuesdays at 4, we can make sure your itinerary gets you there. Similarly, having good data about a place is the first step in making personalized-recommendations based on your unique user-profile. For example, people who use wheelchairs have wildly varying access needs and tailoring recommendations to their unique needs could be the difference between an okay recommendation and a great recommendation.

Understanding when a hotel is an excellent match for one wheelchair user, but a bad recommendation for a different wheelchair user requires detailed accessibility data. Having a travel database enriched with accessibility content also allows us to do “destination discovery” that is aware of your disability. Maybe you hadn’t considered Torredembarra, Spain for your next holiday, but we put it on your radar because it has scuba diving that meets your accessibility needs. Creating this database would not just be a great resource for disabled travellers, but could also be the foundation of other technologies.

Planning with Disabilities in Mind: Incorporating Disability Studies into Mobi.AI's Technology

Tesla Wells explains how Disability Studies helps Mobi reimagine our engineering design process and create products that are useful to everyone.

Read article

Image of rainbow mountains in Cusco, Peru

The Magic in Mobi's Data

Mobi uses an exceptional collection of data—tens of millions of curated points of interest—to generate itineraries anchored in the realities of our constantly changing world, providing a magical experience for our partners and their customers.

Read article