Outcome

Overview

The pilot showed that anonymized e-scooter telemetry can do more than support isolated tagging decisions. Wheels4Wheels developed an automated pipeline that turns an entire ride into structured surface quality proposals for the OpenStreetMap ways it passes through. What previously required manual lookup, selection, and tagging of individual ways can now be processed systematically.

Objectives

The project explored whether anonymized e-scooter telemetry could provide reliable wheelchair accessibility information for OpenStreetMap. The investigation focused on three main goals, and each goal maps to one part of the pipeline that the pilot then put to the test.

The goals translated into a working pipeline over the course of the pilot. The components below were built and tested against Tallinn ride data, covering everything from raw telemetry processing to the review outputs a contributor works with.

Findings

All findings on this page rest on data from the Tallinn pilot. Figures reflect unique, usable rides only, with duplicates and flawed inputs removed. The figures show three key results: Enough data was collected to get robust results. A significant amount of new surface quality information was produced. There's a high match with verified OSM data, effectively validating the methodology. To comply with OSM guidelines, pilot proposals were subject to especially conservative rules. Now that the method is validated, these measures can be gradually relaxed. All data added to OSM can be found here.

Key Learnings

The pilot produced three learnings that carry beyond Tallinn. They concern the quality of the sensor signal itself, the effect of scoring for wheelchair relevance, and how the results fit into OpenStreetMap's existing data model.

The Signal Is Usable

Z-axis standard deviation is a strong indicator of surface roughness, and GPS accuracy, after filtering, is sufficient to align rides with OSM ways. The correlation held across surface types, which is what makes threshold-based classification workable at all.

Wheelchair Perspective Matters

Confidence scores drop sharply during stops and at very high speeds, which matches wheelchair use rather than e-scooter performance alone. Even a single well-recorded ride can provide useful information if confidence and coverage are high enough.

OSM Can Absorb the Data

Calibrated thresholds reproduce existing smoothness=* tags with good agreement, and working at the way level with coverage thresholds fits OSM’s data model without over-fragmenting streets. No schema changes or new tag conventions are required.

Conclusion

The exploration phase confirmed that e-scooter telemetry can provide meaningful wheelchair accessibility insights. The pipeline from raw rides to OSM-ready proposals is reproducible and, with expanded data and community engagement, can scale to improve accessibility mapping globally.