OmniPath: Multi-Modal Framework Audits Wheelchair Accessibility

OmniPath is a system that fuses OpenStreetMap network topology with submeter high-density aerial LiDAR from USGS 3DEP to build a high-fidelity 3D model for auditing wheelchair accessibility. The paper was submitted to arXiv on 23 Jun 2026, and the authors validated the framework against 200 physical ground truth field surveys across the National Mall.

How does OmniPath work?

OmniPath combines OSM network topology and USGS 3DEP LiDAR to create a detailed 3D pedestrian environment, then virtually traverses that network in 0.5 meter increments to analyze surface conditions, quantify slope, cross slope, and vertical discontinuities, and compute a weighted severity score that labels hazards from Mild to Critical. The traversal is agentic in that it performs automated micro-scale inspection rather than returning only a route.

In practice the system converts the fused data into a high-fidelity 3D model, inspects the pedestrian surface at submeter resolution, and measures three physical friction points: running slope, cross slope, and vertical discontinuities. Each measured attribute is compared against ADA compliance standards and combined into a single weighted severity score. The output moves mapping from a binary path presence toward explicit accessibility information about how the surface will feel to a wheelchair user.

How accurate is OmniPath?

The framework showed strong diagnostic reliability for high-severity hazards: the paper reports F1-scores of 0.60 for Severe and 0.58 for Critical categories, validated using stratified random sampling against 200 ground truth field surveys on the National Mall. Those figures are the specific, source-attributed performance metrics the authors highlight for high-severity conditions.

The validation focused on micro-scale inspection accuracy rather than end-to-end routing outcomes. The paper frames these F1-scores as evidence that automated inspection can identify the "invisible" barriers standard maps miss. The authors also note the system was tested with stratified random sampling across a defined testbed, the National Mall, to establish real-world reliability for those severity bands.

Why it matters

OmniPath converts passive mapping data into proactive accessibility audits by adding physical-surface metrics to topology. That matters because platforms like OpenStreetMap frequently show where a path exists but not how it feels to travel on it, a gap that can leave wheelchair users with unreliable route expectations. Automating submeter inspections can scale micro-scale assessments that are otherwise costly to gather by hand.

For planners and accessibility advocates, a reproducible pipeline that quantifies running slope, cross slope, and vertical discontinuities against ADA standards provides a comparable metric set to prioritize repairs. For developers of navigation and mapping services, it offers a path to include hazard severity labels rather than only pass/fail accessibility tags.

What to watch

Watch for the paper's peer-review outcome at IEEE COMPSAC 2026, where the authors submitted the 10-page, 13-figure manuscript. Also watch whether future validations extend beyond the National Mall testbed and whether the approach maintains similar F1 performance when applied in different urban contexts or with other LiDAR sources.

References and concrete points drawn from the paper: the system inspects the network at 0.5 meter increments; it was validated on 200 physical ground truth surveys; and it reported F1 0.60 (Severe) and 0.58 (Critical) for high-severity hazard detection.