Wherobots Mobility Solution Accelerator

A production-ready solution accelerator for processing GPS mobility data using Wherobots and Apache Sedona. Demonstrates a full medallion architecture (Bronze / Silver / Gold) with advanced 3D/4D geometry functions, trajectory construction, map matching, and multi-view analytics output.

Architecture

Raw .plt files ──► [prepare_geolife.py] ──► CSV on S3
                                              │
                    ┌─────────────────────────┘
                    ▼
              ┌───────────┐
              │  01_Bronze │  Ingest raw GPS → 2D point geometries → GeoParquet
              └─────┬─────┘
                    ▼
              ┌───────────┐
              │  02_Silver │  Clean → 4D XYZM points → Trajectories → Map Match
              └─────┬─────┘
                    ▼
              ┌───────────┐
              │  03_Gold   │  H3 heatmaps │ Trajectory explorer │ Clustering
              └───────────┘

Dataset

GeoLife GPS Trajectories (Microsoft Research)

• 17,621 trajectories from 182 users in Beijing (2007–2012)
• Raw GPS points with latitude, longitude, altitude, and timestamps
• Ideal for 3D/4D geometry processing (XYZM)
• Download instructions

Key Features

Feature	Sedona Functions Used
4D Point Construction	ST_MakePoint(x, y, z, m) with elevation + timestamp
Trajectory Building	ST_MakeLine(), ST_IsValidTrajectory()
3D Analysis	ST_Z(), ST_M(), ST_ZMin/Max(), ST_3DDistance()
Elevation Profiles	ST_LineInterpolatePoint(), ST_Z()
Spatial Indexing	ST_H3CellIDs(), ST_H3ToGeom(), ST_GeoHash()
Map Matching	wherobots.matcher.load_osm(), matcher.match()
Stop Detection	dbscan() clustering on low-speed points
Visualization	SedonaKepler.create_map() interactive maps

Notebooks

01 — Bronze: Ingestion & Profiling

• Loads raw GeoLife CSV from S3
• Creates 2D point geometries
• Data quality profiling (null checks, altitude distribution, spatial extent)
• Writes raw GeoParquet

02 — Silver: Transformation & Map Matching

• Cleans invalid records, converts altitude ft → meters
• Constructs 4D XYZM geometries with elevation and epoch timestamp
• Segments trips by time gaps, builds trajectory LineStrings
• Calculates speed, elevation gain/loss, bearing
• H3 and GeoHash spatial indexing
• Map matches GPS traces to OSM road network via WherobotsAI

03 — Gold: Analytics & Visualization

• Analytical View — H3 hexbin heatmaps, temporal patterns, trip statistics
• Exploratory View — Individual trajectories, elevation profiles, matched vs raw routes
• Deep Dive View — DBSCAN stop clusters, hotspots, anomaly detection, road segment analysis

Quick Start

Prerequisites

• Wherobots Cloud account
• Python 3.9+ (for the data preparation script)

Setup

1. Download the GeoLife dataset — see data/README.md

2. Prepare the data (run locally):

   pip install pandas
   python scripts/prepare_geolife.py --input data/Geolife_Trajectories_1.3/Data --output data/geolife_combined.csv

3. Upload to S3:

   aws s3 cp data/geolife_combined.csv s3://<YOUR_BUCKET>/mobility/raw/geolife_combined.csv

4. Run notebooks in Wherobots Cloud in order:

   • 01_bronze_ingestion.ipynb
   • 02_silver_transformation.ipynb
   • 03_gold_analytics.ipynb

5. Configure S3 paths — update the S3_BASE variable in each notebook to point to your bucket.

For Map Matching (Notebook 02)

Download a Beijing OSM extract and upload to S3:

# Download Beijing OSM data from Geofabrik
wget https://download.geofabrik.de/asia/china-latest.osm.pbf
# Or use a smaller Beijing-specific extract
aws s3 cp osm_beijing.xml s3://<YOUR_BUCKET>/mobility/data/osm_beijing.xml

Output Formats

All gold-layer outputs are written as GeoParquet files, compatible with:

• Kepler.gl — drag-and-drop GeoParquet support
• QGIS — native GeoParquet reader
• DuckDB Spatial — query GeoParquet with SQL
• Felt — upload GeoParquet directly
• Foursquare Studio — cloud visualization

License

This solution accelerator is provided as-is for educational and demonstration purposes. The GeoLife dataset is provided by Microsoft Research under their terms of use.