Skip to content

Map-Localization Release Gates for Timing Health

Last updated: 2026-05-09

Purpose

This page defines release gates that tie map publication, localization release, and fleet rollout to timing health. A localization stack can pass nominal accuracy tests and still be unsafe if PTP, sensor timestamps, replay semantics, TF filters, or runtime latency are unverified. A map can pass geometric QA and still be unsafe if its source sessions were built from skewed or stale sensor data. These gates prevent timing defects from entering vehicle release evidence or permanent map layers.

This page consolidates evidence from time-sync fault injection, timestamp shift sweep, sensor dropout, latency, and jitter stress, replay time semantics validation, SLAM integrity under timing errors, SLAM map benchmark protocol, and the SLAM timing health dashboard.

Release Scope

Release itemTiming evidence required
Localization softwarePTP/PHC health, timestamp sweep margin, dropout/jitter stress, integrity under timing faults
Sensor driver or firmwareTimestamp source declaration, sensor-native clock mode tests, fallback detection
Map packageSource-session timing health, map-building replay validity, tile-level timing provenance
Calibration packageTime offset/extrinsic coupling evidence and calibration run timing health
Replay datasetROS time, /clock, TF, message-filter, MCAP/rosbag metadata validation
Fleet rolloutDashboard coverage, alert thresholds, operational response, canary timing envelope

Required Setup

ItemRequirement
Release manifestBuild, map, calibration, sensor manifest, PTP profile, timing thresholds, ODD scope
Timing evidence bundleFault injection, timestamp sweep, latency/jitter, replay semantics, and integrity reports
Map provenanceSource session IDs, tile IDs, map builder version, timing health tags, quarantine state
Fleet observabilitySLAM timing health dashboard, alert routes, incident joins, and canary cohort labels
Owner approvalsV&V timing, perception/SLAM, mapping, runtime/platform, fleet operations, and safety sign-off

Pass and Block Gates

GateEvidencePass conditionBlock condition
MLT0 timing provenanceSensor manifest, PTP profile, PHC mapping, GNSS/PPS wiring, driver configEvery timing source and clock domain is declared and versionedUnknown timestamp source, mixed epochs, or undocumented fallback
MLT1 nominal timing healthPTP/PHC logs, sensor stamp age, inter-sensor skew, topic stateNominal run stays inside approved green band by sensor and ODD sliceOffset/skew exceeds envelope without alert
MLT2 fault detectionTime-sync fault injection reportPTP loss, failover, drift, path asymmetry, and timestamp fallback are detected within budgetTiming fault remains silent to runtime and fleet telemetry
MLT3 timestamp marginTimestamp shift sweep reportMonitor/degrade threshold is below unsafe shift with approved marginNormal timing variation overlaps silent-failure region
MLT4 latency and jitterDropout/latency/jitter stress reportQueues, filters, and stale-data monitors fail closed under stressStale/future data consumed as current safety input
MLT5 replay validityReplay time semantics reportRelease replays preserve /clock, TF, message filters, message order, and MCAP/rosbag timing metadataInvalid replay used for candidate selection or release claim
MLT6 SLAM integrityIntegrity report under timing errorsPose bound contains truth or alert triggers before unsafe exposureHazardously misleading localization
MLT7 map-source timingMap source-session timing tags and QA reportEvery source traversal for published tile is timing-green or explicitly reviewedTiming-degraded session contributes to permanent map without quarantine
MLT8 canary observabilityFleet dashboard and alert dry runTiming health panels, alerts, and runbooks are live before rolloutCanary cannot detect or triage timing health regression
MLT9 incident evidenceData contract and retention checkTiming telemetry joins to bag/MCAP, map, build, calibration, and incident IDsSafety event cannot be reconstructed due to missing timing evidence

Metrics Required in Release Packet

Metric groupRequired fields
Clock disciplineGrandmaster ID, PTP state, PHC offset, system-to-PHC offset, path delay, frequency correction
Sensor timingTimestamp source, stamp age, inter-arrival jitter, inter-sensor skew, fallback status
ROS/replay timing/clock policy, record/log/publish time, message order, replay determinism
TF and filtersTF lookup failures, cache bounds, message-filter match/drop reasons, queue age
Runtime latencySource stamp to localization/perception output, p50/p95/p99/p99.9, burst max
Localization integrityPose error, covariance/protection level, residuals, NEES/NIS, time-to-alert
Map provenanceSource session IDs, timing health tags, tile IDs, map build time, quarantine/review state
OperationsAlert counts, route/ODD slice, vehicle ID, operator response, maintenance ticket, rollback/quarantine action

Decision Rules

DecisionRule
Software release passAll gates MLT0 through MLT6 pass for intended ODD slices; MLT8 and MLT9 active for canary
Map publication passMLT0, MLT1, MLT5, MLT7, and map QA gates pass for every source traversal and tile
Calibration release passCalibration run timing health is green and timestamp offset/extrinsic coupling is documented
Canary onlyMinor timing margin gap has approved mitigation, dashboard is active, and route/ODD is restricted
BlockAny hazardously misleading localization, stale safety input, invalid replay evidence, or timing-corrupted map source
InconclusiveEvidence is missing, replay semantics are invalid, or test slice coverage is insufficient

Operational Response

Release findingRequired action
PTP/PHC margin below targetHold release or restrict ODD; fix clock source, network path, or monitor threshold
Sensor timestamp fallback not detectedBlock affected sensor driver/firmware release
Replay semantics invalidRe-run release metrics after replay harness correction; invalidate prior timing-sensitive claims
SLAM integrity unsafe without alertBlock release and require monitor/fusion/covariance redesign
Map source session timing-degradedQuarantine affected tile or rebuild from timing-healthy source traversals
Canary timing alert spikePause rollout, route hold if safety-critical, compare against validation envelope
Missing incident timing fieldsBlock canary expansion until data contract and dashboard joins are fixed

Evidence Package Checklist

ArtifactRequired content
Timing release manifestBuild/map/calibration IDs, sensor timing modes, PTP profile, thresholds, ODD scope
Fault injection reportTime-sync fault matrix, monitor response, recovery, safety actions
Timestamp sweep reportResponse curves, silent-failure analysis, release margins
Latency/jitter reportTopic rates, queue age, filter drops, stale-data rejects, runtime latency percentiles
Replay validation report/clock, TF, message filters, rosbag2/MCAP timing semantics, determinism
Integrity reportProtection-level containment, alert timing, unsafe-without-alert intervals
Map timing provenanceSource traversal timing health, tile quarantine/review decisions
Dashboard readinessPanels, alert thresholds, runbook links, incident routing, owner acknowledgements

Owner Handoffs

OwnerResponsibility
Release managerEnforce MLT gates before software, map, calibration, or canary promotion
V&V timing leadOwn fault injection, timestamp sweep, latency/jitter, and replay validation reports
Perception/SLAM ownerFix fusion/localization defects and define timing margins
Mapping ownerEnforce timing provenance and map-source quarantine
Runtime/platform ownerMaintain PTP/PHC, ROS time, TF/filter telemetry, and alert implementation
Fleet operationsMonitor canary timing health and execute route holds/maintenance workflows
Safety leadApprove alert limits, residual risk, ODD restrictions, and release disposition

Sources

Public research notes collected from public sources.