Data Gathering

Overview

Anonymized telemetry from Bolt e-scooter rides is combined with manual mapping to produce a structured dataset for wheelchair accessibility. It captures accelerometer, GPS, and speed measurements across areas in Tallinn and is validated through ground-truth observations.

Objectives

Data collection set out to fill the gaps in wheelchair accessibility mapping by cross-referencing human input with sensor data from e-scooters, so the two could validate one another. The work focused on four main goals.

Collection Goals

Methodology: Compare human input with e-scooter sensor data

Prototype Map: Highlight accessible and non-accessible areas

Data Gaps: Identify for prioritization in future mapping efforts

Foundation: Make findings available open source to map developers

Telemetry Source

Anonymized e-scooter telemetry data is collected from Bolt e-scooter rides. Each ride generates a JSON file containing high-frequency sensor measurements, enabling detailed analysis of surface conditions for wheelchair accessibility mapping.

Ride Telemetry Sample

{
"rideData": [
  {
    "timestamp": 1234567890000,
    "accelerationCoord": { "x": 120, "y": -50, "z": 981 },
    "gpsPos": { "lat": 59.4372, "lon": 24.7453 },
    "currentSpeed": 278,
    "gpsInfo": { "date": "2024-01-15", "time": "14:30:25" }
  }
]
}

{"rideData": [  {    "timestamp": 1234567890000,    "accelerationCoord": { "x": 120, "y": -50, "z": 981 },    "gpsPos": { "lat": 59.4372, "lon": 24.7453 },    "currentSpeed": 278,    "gpsInfo": { "date": "2024-01-15", "time": "14:30:25" }  }]}

Key Parameters

Sensor	Details
Accelerometer	Frequency: 30Hz (30 samples/sec) Axes: X, Y, Z in milligravity (1/1000 G) Primary metric: Z-axis standard deviation (surface roughness indicator)
GPS	Frequency: 1Hz (1 sample/sec) Coordinates: WGS84 latitude/longitude Accuracy: ±3–5 meters in urban environments
Speed	Unit: 1/10 km/h (raw telemetry) Conversion: raw value ÷ 10 → km/h Optimal range for wheelchair assessment: 2–6 km/h
Manual Validation	Structured on-the-ground mapping using StreetComplete app for surface types, ramps, curbs, etc. (for ground-truth and calibration)

Manual Collection

StreetComplete is a mobile app that enables volunteers to contribute structured data to OpenStreetMap in a simple and gamified manner. The app presents “Quests”, which are small, focused questions about streets and infrastructure. Each Quest corresponds to a specific OSM tag, ensuring that volunteers’ answers are standardized and directly usable in the map database.

StreetComplete Features

Key Feature	Description
Focused Attributes	Each Quest targets a single mapping attribute, such as surface type, ramp presence, or curb height
Consistent Input	Volunteers select from predefined options, ensuring high-quality, uniform tagging
Automatic Integration	Answers are converted directly into OSM key=value pairs
Easy Contribution	Quests are quick and intuitive, allowing volunteers to participate without deep OSM knowledge

Selected Quests

Topic	Quest	OSM Code
Sidewalk Existence	Is there a sidewalk along this street?	`sidewalk=*`
Sidewalk Surface	What’s the surface here?	`surface=` (or `sidewalk:surface=`)
Surface Quality	What’s the surface quality here?	`smoothness=*`
Ramps	Do these steps have a ramp? What kind?	`highway=steps`
Street Curbs	What’s the height of this curb?	`barrier=kerb`, `footway=sidewalk`, `footway=crossing`
Crossing	Is there a crossing here?	`footway=crossing`
Crossing Island	Does this crossing have an island?	`crossing:island=yes`

Location

Before selecting specific areas for data collection, several criteria guided the choice of candidate cities and neighborhoods for mapping accessibility.

Candidate Cities

City	Population	Area (km²)	Characteristics
Oslo	724K	480 km²	High accessibility awareness and strong pedestrian infrastructure
Stuttgart	620K	208 km²†	Dense urban layout with mixed sidewalks and streets Good test for complex mapping
Berlin	3.9M	891.7 km²‡	Large city with varied terrain and extensive wheelchair‑accessibility needs
Madrid	3.3M	604.3 km²	Mediterranean streets with historical cobblestones and urban obstacles
Tallinn	456K	159 km²§	Mixed cobblestones and smooth urban zones Bolt HQ city E-scooters allowed on sidewalks at pedestrian speed

* Municipality-level population and area figures from public statistics, rounded; indicative for site selection

† Stuttgart’s inner‑city area estimate

‡ Berlin covers ~891.7 km² as municipality

§ Tallinn municipality area approximate

Ultimately, Tallinn was chosen because Estonian law permits e-scooters on sidewalks at pedestrian speed, and the city offered a controlled, representative environment for testing the methodology. Within Tallinn, certain areas were prioritized for logistical reasons.

Tallinn Areas

Area	Details
Kadriorg	A historic and residential area known for its parks and the Kadriorg Palace
Telliskivi	A creative district with a mix of industrial and modern architecture, popular among locals and tourists
Kalamaja	A bohemian neighborhood characterized by wooden houses and proximity to the sea
Balti Jaam	The area around the main railway station, a hub for transportation and commerce
City Center	The central business district largely including pedestrian zones

Data Collection

To ensure the accuracy and reliability of the data collected by e-scooters, the Wheels4Wheels team conducted a volunteer validation phase. Human contributors were enlisted to manually verify accessibility features in the city, providing a ground-truth reference against which sensor measurements could be compared. This made it possible to cross-check the automated readings and refine the mapping parameters before wider deployment.

Experiment Setup

Item	Details
Participants	10 volunteers participated in the experiment
Duration	10 working days of data collection
Methodology	Volunteers used the StreetComplete app to manually map accessibility information, including surface type, ramps, presence of sidewalks, and obstacles
Cross-validation	Data collected by e-scooters (accelerometer, gyroscope, GPS, and cameras) was cross-referenced with the manual input to verify accuracy and ensure high-quality mapping

Data Extraction

The data extracted from e-scooter telemetry undergoes initial processing to isolate relevant measurements. This step ensures that only the information necessary for assessing surface quality, inclines, and accessibility features is retained, providing a clean and structured foundation for subsequent analysis.

Primary Metrics

Metric	Description
z_std	Standard deviation of Z-axis acceleration (mg units)
speed_kmh	Speed in kilometers per hour
Coordinates	GPS latitude/longitude pairs
Timestamp	Unix timestamp for synchronization

Raw metrics only become useful once they are condensed. Three derived quantities are computed from the stream and carried through the rest of the pipeline, each summarizing one aspect a routing engine ultimately cares about.

Telemetry Derivatives

Surface Roughness

Calculated from z_std over 1-second windows

Movement State

Stationary vs. moving on speed threshold

Data Quality

GPS accuracy, accelerometer stability

Time Series Visualization

The chart below mirrors the full-ride time series from the surface quality analysis. Flat segments with very low z_std correspond to idle periods (speed ≈ 0), while high spikes indicate rough surfaces under movement.

Loading chart data...