Validation
What we measure, how we measure it.
Autonomy that makes life-safety decisions can only be as trustworthy as the sensor beneath it — so this page lays out how we hold our own sensing to account.
This page is intentionally non‑confidential. It describes a validation mindset: define KPIs, measure them under controlled conditions, then confirm in realistic scenes.
Core KPIs (examples)
Range & precision
Distance accuracy and stability
- Targets: calibrated reflectors and diffuse panels
- Metrics: bias, σ, tail risk
- Conditions: sunlight, fog/rain proxies, angle
Latency
Sensor-to-decision time
- Frame interval and per‑ROI update cadence
- End-to-end timing budget (sensor → compute)
- Jitter and worst-case behavior
Coexistence
Interference resilience
- Multi‑LiDAR traffic injection tests
- False positives / dropouts / artifacts
- Mitigation impact on throughput
Reporting (what “good” looks like)
Repeatability
Same setup → same outcome
We track calibration state, ambient conditions, and test scripts so results are comparable over time.
Traceability
From claim → test → evidence
Each externally stated KPI ties to a test method and a reporting artifact.
Honest bounds
Disclose constraints
We report where performance degrades (weather, reflectivity, speed) and how mitigations trade off.