Static Asset Preservation in Map Cleaning

Last updated: 2026-05-09

Dynamic-object removal is only acceptable when it preserves the static structure that localization, planning, geofencing, and operator review depend on. A cleaner that removes parked vehicles but also erodes curbs, stand markings, terminal edges, poles, gates, or docking aids has reduced map clutter by creating a safety defect.

This page defines the static-preservation validation case for airside LiDAR map cleaning.

Preservation Claim

For the approved airside ODD and map layer, the map-cleaning pipeline removes transient dynamic clutter while preserving surveyed or repeatedly observed static assets that are required for localization, semantic map alignment, operational boundaries, and safety review.

The claim is limited to the sensor rig, map resolution, cleaner version, route/stand set, weather slices, and review process in the release evidence package.

Asset Classification

Asset class	Examples	Map treatment	Review owner
Permanent static	Terminal walls, jet bridge fixed bases, light poles, hydrants, curbs, surveyed stand markings	Preserve in localization or semantic layer	Mapping lead
Safety static	stop bars, hold lines, restricted-area paint, service-road boundaries	Preserve and cross-check against vector map	Safety and airport ops
Movable-static	cones, chocks, barriers, dollies, belt loaders, parked buses, staged carts	Keep out of permanent layer unless approved as an overlay	Map operations
Current dynamic	aircraft, tugs, people, moving GSE, temporary work crew	Remove from static layer; retain as replay evidence	Perception and ops
FOD/hazard	debris, loose hardware, straps, pavement fragments	Do not erase as noise; preserve as alert evidence	Safety and FOD owner
Unknown/review	occluded, sparse, low-confidence, or disagreement regions	Quarantine tile or require human review	Release approver

Test Matrix

Static feature	False deletion mode	Required test slice	Evidence to keep
Stand and service-road markings	Removed as low-height ground clutter	wet/dry, day/night, worn paint	before/after point cloud, intensity image, line residual
Curbs and wheel stops	Removed by terrain or ground filters	shallow incidence, sparse range, near aircraft stand	cross-section profile and retained-point ratio
Poles and signs	Removed as isolated outliers	open apron, terminal edge, reflective surfaces	object-level retention and localization impact
Docking aids and fixed cabinets	Classified as parked equipment	aircraft present/absent paired sessions	asset ID, bounding volume, reviewer decision
Terminal glass/edges	Over-cleaned due to multipath	sun glare, night lighting, wet surface	raw scans, rejected points, ghost/real label
Low fixtures	Removed as FOD/noise	chocks nearby, hoses, drains, covers	semantic label and close-range inspection

Metrics

Metric	Definition	Gate use
Static preservation rate	retained ground-truth static points divided by all ground-truth static points	Primary preservation metric by asset class
Static erosion by asset	false-deleted points or cells inside an approved static asset volume	Blocks publication for safety-critical assets
Localization delta	change in NDT/ICP residuals, inliers, covariance, relocalization success, ATE/RPE	Cleaned map must not degrade localization in release zones
Semantic alignment delta	distance between retained point cloud features and Lanelet2/vector features	Protects stop lines, boundaries, stands, and route graph
Removed-layer review rate	percentage of rejected points accepted as legitimate deletion by human review	Detects overly aggressive cleaners
Quarantine rate	percentage of tiles requiring unknown/review disposition	Measures operational scalability

Publication Gates

Gate	Pass condition	Blocker
Source provenance	raw scans, poses, calibration, map version, cleaner config, and rejected layer are archived	missing rejected points or untraceable cleaner version
Ground-truth coverage	every safety-critical asset class has labels in affected zones	only aggregate PR/RR metrics are available
Localization replay	candidate map is neutral or better on residuals, inliers, covariance, and recovery	scan matching worsens in any approved route segment
Semantic validation	vector map and point cloud agree within the site tolerance	disconnected, shifted, or stale semantic features
Human review	false deletions are resolved or tile is quarantined	reviewer cannot inspect before/after evidence
Rollback	prior map bundle and compatible config remain deployable	no signed rollback target

Release Checklist

1. Freeze the raw survey bundle and cleaner configuration before running acceptance metrics.
2. Score static preservation per asset class, not only aggregate map-cleaning score.
3. Compare at least one conservative baseline and one production candidate on the same logs.
4. Review the rejected layer around stands, gates, pedestrian paths, and service-road boundaries.
5. Run localization replay on raw, cleaned, and candidate-over-cleaned maps.
6. Keep movable-static assets out of the permanent map unless a lifecycle policy approves them.
7. Link every accepted false deletion to a reviewer decision or a non-safety artifact class.
8. Reject or quarantine any tile with unresolved erosion of safety-critical static structure.

Sources

• FAA AGVS on Airports: https://www.faa.gov/airports/new_entrants/agvs_on_airports
• FAA Emerging Entrants Bulletin 25-02: https://www.faa.gov/airports/new_entrants/bulletins/25_02
• KTH Dynamic Map Benchmark: https://kth-rpl.github.io/DynamicMap_Benchmark/
• Dynamic Points Removal Benchmark paper: https://arxiv.org/abs/2307.10593
• MapCleaner article: https://www.mdpi.com/2072-4292/14/18/4496
• RI-DVP sparse LiDAR map-cleaning article: https://www.mdpi.com/2072-4292/18/5/821
• Autoware map component design: https://autowarefoundation.github.io/autoware-documentation/main/design/autoware-architecture/map/
• Autoware NDT scan matcher: https://autowarefoundation.github.io/autoware_core/pr-602/localization/autoware_ndt_scan_matcher/
• Local context: ../airside-dynamic-map-cleaning-benchmark.md