OmniPath: A Multi-Modal Agentic Framework for Auditing Wheelchair Accessibility

Traditional digital maps often fail wheelchair users by prioritizing path connectivity over physical surface quality. Standard 2D maps frequently miss "micro-barriers"—such as steep inclines, uneven pavement, or vertical gaps—that render a path impassable for a wheelchair user. This paper asks: Can we move from reactive, human-reported accessibility mapping to a proactive, agentic system that automatically audits the physical environment using high-resolution geospatial data?

OmniPath functions by fusing two distinct data layers: the logical network topology from OpenStreetMap (OSM) and the physical surface topography from high-density aerial LiDAR (USGS 3DEP). The framework employs a virtual agent that traverses the pedestrian network in 0.5-meter increments. At each segment, the agent calculates three critical metrics: running slope, cross slope, and vertical discontinuities. These measurements are compared against ADA compliance standards and aggregated into a Weighted Severity Score, which categorizes segments from "Mild" to "Critical" based on the biomechanical difficulty they pose to wheelchair users.

The researchers validated the OmniPath framework against 200 physical ground-truth field surveys conducted across the National Mall. The system demonstrated strong diagnostic reliability in identifying high-severity hazards, achieving F1-scores of 0.60 for "Severe" and 0.58 for "Critical" categories. By automating this micro-scale inspection, the framework successfully identifies physical barriers that are invisible to standard satellite imagery and 2D vector maps, allowing for the generation of accessibility-aware routes before a user begins their journey.