AccessMapAI

AccessMap AI builds a personalized, explainable map of the city for people with different accessibility needs. It routes around the things other maps pretend aren't there.

Comment

AccessMap AI

AI-Powered Accessibility Intelligence Platform

HackDavis 2026

Profile-aware, hazard-aware, explainable navigation for accessible urban mobility.

Inspiration

It started during a Tuesday afternoon lecture in AMS 012 — US Disability and Culture.

Adi was introduced to the social model of disability, a framework built around one powerful idea:

A person is not disabled by their body. They are disabled by an environment that was never designed for them.

After that lecture, everyday infrastructure failures suddenly became impossible to ignore:

  • stairs without ramps
  • crosswalks without audible signals
  • sidewalks that abruptly end
  • inaccessible entrances hidden behind “main” entrances

Within days, we began cataloging accessibility barriers across the UC Davis campus — places we walked through every day that others physically could not.

Then we looked at the tools people actually use.

Google Maps can reroute around traffic.
It cannot reroute around a missing curb ramp.

Apple Maps exposes a single wheelchair=yes tag and stops there.

Meanwhile, OpenStreetMap already contains rich accessibility metadata:

  • slope
  • sidewalk surface
  • kerb quality
  • tactile paving
  • crossing signals
  • lighting coverage

Yet almost none of this information is meaningfully used by mainstream navigation systems.

1 in 4 adults in the United States lives with a disability.
That is over 70 million people navigating environments designed primarily for everyone else.

The disconnect between accessibility research and real-world navigation technology became the foundation for AccessMap AI.

“Most navigation apps optimize for speed — not accessibility.
A missing curb ramp can make an entire route impossible.”

That became the project’s north star.

What AccessMap AI Does

AccessMap AI is a profile-aware, hazard-aware, explainable accessibility map.

Users can:

Select accessibility profiles

  • wheelchair users
  • blind / low-vision users
  • elderly users
  • neurodivergent users
  • temporary injury accommodations

Generate accessibility-optimized routes

Instead of simply finding the shortest route, AccessMap AI prioritizes the most passable and accessible path for the user’s needs.

View detailed route scoring

Each route includes explainable score breakdowns for:

  • slope
  • surface quality
  • lighting
  • hazards
  • crossing accessibility

Read natural-language explanations

The system explains why a route was selected and what tradeoffs influenced the result.

Report live hazards

Users can submit hazards such as:

  • blocked sidewalks
  • broken curb ramps
  • construction zones

Routes dynamically adapt around reported hazards in real time.

Analyze sidewalks using AI

Users can upload sidewalk photos for accessibility analysis powered by Gemini.

Use spoken navigation

Turn-by-turn directions can be read aloud through Web Speech integration for blind and low-vision users.

System Architecture

Data Foundation

The routing graph is constructed entirely from open-source geospatial data.

Source Usage
OpenStreetMap (Overpass) Sidewalks, crossings, kerbs, tactile paving, lighting, roads, buildings
Open-Meteo Elevation API Slope estimation using elevation gradients
Unitrans GTFS UC Davis bus routes and stops

The preprocessing pipeline generates an enriched NetworkX graph containing approximately:

  • 19,700 nodes
  • 20,700 edges

Each edge stores derived accessibility features such as:

  • surface roughness
  • slope severity
  • kerb quality
  • tactile coverage
  • lighting density
  • crossing signal availability

Routing Engine

Each accessibility profile defines a weighted sensitivity vector representing how important different accessibility factors are for that user.

The routing engine evaluates:

  • distance
  • slope
  • surface quality
  • hazards
  • crowd density
  • noise
  • accessibility constraints

The route cost is calculated by combining edge distance with weighted accessibility penalties. Each environmental factor (such as slope or surface roughness) contributes to the total path cost depending on how important that factor is for the selected accessibility profile.

Hard accessibility constraints such as:

  • avoiding stairs
  • slope limits
  • minimum width requirements
  • avoiding unpaved paths

are encoded as large penalties, allowing Dijkstra’s algorithm to naturally avoid inaccessible routes.

For combined accessibility profiles, the system uses the most restrictive constraints and highest factor sensitivities from each selected profile to ensure accessibility needs are preserved.

Hazard-Aware Scoring

User-reported hazards are stored dynamically in Supabase and integrated directly into route scoring.

For each hazard, the system computes the minimum geographic distance between the hazard and the route. Accessibility penalties are then assigned based on proximity:

  • hazards within 15 meters produce severe penalties
  • hazards within 30 meters produce moderate penalties
  • hazards within 60 meters produce minor penalties

These penalties are aggregated into the overall accessibility score while preventing a single area from completely invalidating the route.

This creates:

  • real-time rerouting
  • transparent hazard scoring
  • explainable accessibility penalties

Users can clearly see:

  • which hazards affected the route
  • how close they are
  • how much they lowered the score

Crowd and Noise Modeling

Real cities rarely provide live accessibility-aware crowd data.

To estimate accessibility-relevant congestion, AccessMap AI models:

  • time-of-day activity
  • building density
  • proximity to campus hotspots

The crowd model combines these signals into a single accessibility-aware congestion score.

Noise levels are estimated using road classification and distance-based attenuation models inspired by FHWA traffic-noise research.

--

Gemini Sidewalk Analysis

Users can upload sidewalk photos for AI-powered accessibility analysis.

Gemini evaluates:

  • obstruction severity
  • surface quality
  • slope
  • wheelchair accessibility

The backend applies additional normalization logic to correct inconsistent scoring behavior and improve reliability.

A mock mode allows offline demos without consuming API quota.

Frontend Stack

  • Next.js 15 App Router
  • TypeScript
  • Tailwind CSS v4
  • shadcn/ui
  • Mapbox GL JS
  • Supabase Authentication
  • Web Speech API

The UI emphasizes:

  • explainability
  • accessibility-first interaction
  • persistent route visibility
  • real-time hazard awareness

Major Challenges

Combined Profile Hazard Logic

Hazard filtering originally failed for multi-profile users because hazards were matched against a combined profile string rather than intersecting profile sets.

The fix required propagating original profile vectors directly into the routing cost layer.

Invisible Hazard Penalties

Routes near hazards still produced unrealistically high scores because static edge scoring ignored live hazards entirely.

We solved this using post-route hazard distance scanning and live penalty aggregation.

Gemini Score Inversion

Gemini occasionally interpreted accessibility scores as “problem severity” rather than “quality.”

A stricter rubric and post-processing sanity checks corrected the issue.

Mapbox Styling Conflicts

Mapbox popup styling conflicted with shadcn dark mode theming due to stylesheet specificity issues.

Custom Tailwind selectors and explicit popup overrides resolved the problem.

What We Learned

Accessibility is Deeply Multidimensional

Accessibility cannot be reduced to a single boolean tag.

True accessibility involves:

  • slope
  • surface material
  • kerb geometry
  • lighting
  • noise
  • crossing safety
  • environmental predictability

Static Maps Are Not Enough

Accessibility changes dynamically.

Construction, flooding, broken ramps, and temporary obstructions require real-time community feedback systems.

Explainability Creates Trust

Users need to understand:

  • why a route was selected
  • which constraints mattered
  • which hazards affected the score

Transparency is essential for accessibility-critical systems.

The Social Model Becomes Software Architecture

Once disability is understood as a mismatch between person and environment, the routing system naturally becomes an environmental reasoning engine.

The mathematics simply operationalizes the social model into infrastructure-aware optimization.

Future Work

  • Hazard expiration and community verification
  • Multi-city onboarding pipelines
  • Indoor accessibility routing using indoor=*
  • Accessibility infrastructure heatmaps for city planners
  • Sign-language route explanation videos

Closing

The world was not built equally for everyone.

AccessMap AI aims to make it more understandable, navigable, and accessible — one route at a time.

Built With

Share this project:

Updates