Skip to content

Map Hygiene Operational Monitoring

Last updated: 2026-05-09

Map hygiene needs runtime monitoring after publication. A map that passed offline validation can still fail in operation because the site changed, a temporary overlay expired, a vehicle received the wrong bundle, or perception disagrees with static assumptions.

Monitoring Goals

GoalSignal
detect stale or wrong mapperception-map disagreement, operator reports, repeated route failures
catch localization regressionresiduals, covariance, inlier count, NDT score, relocalization time
catch false-free-space riskcurrent obstacle where map says clear, FOD ticket in route corridor
control temporary overlaysexpiry, owner acknowledgement, active vehicle count
support incident reviewactive map ID, prior map ID, rejected layer, raw evidence links
guide data collectionzones with repeated uncertainty, sparse features, or reviewer burden

Telemetry Fields

FieldTypeNotes
map.site_idstringairport or test-site identifier
map.tile_idstringstable tile/zone/stand identifier
map.bundle_idstringsigned map package ID
map.layer_idsstring arraypoint cloud, semantic, overlay, hazard, unknown
map.cleaner.versionstringalgorithm/model/config version
map.release_stateenumdraft, validation, canary, active, rolled_back, retired
map.hygiene.static_preservation_ratedoubleby zone or asset class
map.hygiene.ghost_ratedoubleresidual dynamic clutter per tile or route
map.hygiene.fod_candidatesintretained/reviewed FOD-like objects
map.hygiene.unknown_area_m2doubleunknown or quarantined area
localization.ndt_scoredoublealign with Autoware NDT debug/diagnostic outputs
localization.covariance_xydouble arraycovariance or derived error ellipse
localization.inlier_ratiodoublescan-to-map health
map.disagreement.countintcurrent perception disagrees with map
map.overlay.expiry_timetimestamprequired for temporary assets
vehicle.active_map_idstringvehicle-reported active bundle

Use OpenTelemetry semantic conventions where they fit, and publish a map-specific schema URL for custom fields so dashboards and offline analysis can handle schema evolution.

Alerts

AlertTriggerAction
map mismatchvehicle active map differs from dispatch expectationstop dispatch or force reload
localization degradationsustained residual/covariance increase by tilecanary pause or rollback
high disagreementrepeated perception-map conflict in same zonecreate map-change ticket
expired overlaytemporary overlay active past expiryblock route or renew approval
FOD conflictFOD/hazard candidate overlaps cleaned/removed layersafety review ticket
unknown growthunknown/quarantine area exceeds route thresholdblock publication or data collection
intervention clusterremote assist/manual takeover by map tileincident triage

Dashboard Views

ViewContents
release healthmap bundle, release state, cohort, rollback readiness
localization by tileNDT score, covariance, residuals, relocalization failures
hygiene by assetstatic preservation, ghost rate, movable-static decisions
FOD and hazardsretained candidates, inspection status, false alarms, closures
overlaysowner, expiry, affected routes, active vehicles
incidentsactive/prior map ID, logs, reviewer records, source evidence

Operational Rules

  1. Every mission log must include the active map bundle ID and active overlay IDs.
  2. Every map-related alert must include tile, route, vehicle, software, sensor config, and timestamp.
  3. Canary promotion requires monitoring coverage for the operational slices affected by the release.
  4. Diagnostics should feed the same incident workflow as software and vehicle health alerts.
  5. Map schema changes must be versioned so older dashboards do not silently misread fields.
  6. Retain telemetry, raw evidence, and map bundles for any incident or safety event.

Sources

Public research notes collected from public sources.