Detector-free LiDAR mapping

Keep moving objects from becoming LiDAR map ghosts.

Accumulated scans can turn cars, pedestrians, and other transient returns into ghost geometry. Clean the map with geometry-only methods, or filter a pose-aligned ROS2 stream — no GPU and no learned detector required.

66.3% moving GT removed in the AV2 audit

97.4% static GT kept in the same proof

0 GPU NumPy-only core; browser demo never uploads data

Try it in your browser See the audited proof Install from GitHub

Try it before installing

Run the real NumPy implementation in the browser. Choose Box, Range, or Temporal, load a public preset, or drop your own point cloud. No upload.

Open the playground → See the audited proof

Same-pose Argoverse 2 accumulation, moving-track ground truth, full point counts, and a separate JSON artifact. The evidence is detector-free and reproducible.

Inspect the AV2 audit → Add it to ROS2

Insert the pose-aware realtime node between your LiDAR frontend and map backend. Timestamped TF, deskewed input, and fail-open behavior are documented.

Open the integration guide →