Production-grade, reproducible Python engineering playbooks for high-definition mapping and autonomous-vehicle spatial data — from raw LiDAR returns and OpenDRIVE schemas to lane graphs, sensor fusion, and fleet-scale validation.
vehiclemapping.org is a deep, practitioner-focused knowledge base for the engineers who build the spatial layer of autonomous driving. Every guide is written for people who copy code straight into real pipelines, so correctness and reproducibility come first: strict coordinate-reference-system governance, sub-decimeter tolerance budgets stated as explicit numbers, memory-bounded batch processing, and safety-gated validation modelled on automotive functional-safety practice.
It exists because HD-mapping and AV spatial work fails in predictable, expensive ways — datum drift leaks
silent error into localization, malformed map schemas reach the planner, point clouds smear under ego-motion,
sensor extrinsics fall out of calibration, and streaming jobs collapse under backpressure. This site collects
the patterns that keep those pipelines honest, each grounded in the production Python toolchain: pyproj,
shapely, open3d, lxml, numpy, scipy, and networkx.
- AV & robotics engineers building localization, mapping, and planning stacks on ROS 2
- HD-mapping specialists working with OpenDRIVE, Lanelet2, and NDS.Live
- Python GIS developers wiring up reproducible geospatial pipelines and CRS-correct transforms
- Automotive tech teams deploying and validating map and sensor-fusion pipelines at fleet scale
| Area | What it covers |
|---|---|
| 🗺️ HD Mapping Architecture & Spatial Data Standards | Coordinate governance and UTM zone handling, OpenDRIVE parsing & XSD validation, lane-level topology and successor graphs, content-addressed version control, and OpenDRIVE ↔ Lanelet2 ↔ NDS.Live interoperability. |
| 🛣️ Lane Geometry Extraction & Road Network Processing | Centerline algorithm selection (midpoint vs Voronoi vs QP), curvature and clothoid fitting, superelevation from point clouds, batch lane-attribute extraction with Dask, topological validation, and routable road-network graphs. |
| 🎯 Sensor Fusion & Spatial Data Alignment | Temporal synchronization and SLERP motion compensation, multi-sensor coordinate alignment with tf2, ICP/NDT/global point-cloud registration, extrinsic calibration for LiDAR/camera/IMU, and asynchronous pipeline architecture. |
- Runnable, production-grade code — every Python example targets a real toolchain, states explicit numeric tolerances (≤0.05 m RMSE, ≤0.1 m lateral error, sub-millisecond timing budgets), and raises on failure rather than degrading silently.
- Correctness where it's hard — axis-order safety, datum and geoid separation, junction topology, rolling-shutter and motion distortion, and calibration-graph consistency are treated as first-class engineering concerns.
- Original diagrams — each guide carries hand-authored, theme-aware inline SVG diagrams (no stock art, no runtime chart libraries) that explain the hardest concept on the page.
- Fast & accessible — a static build that passes automated WCAG 2 A/AA, performance, structured-data, and link-integrity checks before every deploy.
- Eleventy static site generator (Nunjucks + Markdown)
- KaTeX for coordinate-transformation and geometry math
- Hand-authored inline SVG diagrams and a lightweight, dependency-free design system
- Progressive Web App with an offline service worker
- Deployed on Cloudflare Workers
# install dependencies
npm install
# build the static site into _site/
npm run build
# serve locally with live reload
npm run serve
# build and deploy to Cloudflare
npm run deploycontent/ Markdown source for every guide, organized by topic area
src/ Layouts, partials, data, CSS, JS, and static assets
src/index.njk Home page
src/*.njk 404, sitemap, robots, manifest, service worker
- Live site: https://www.vehiclemapping.org
- Source: https://github.com/vehiclemapping/vehiclemapping
Released under the MIT License.