FAQ

製品・データ・導入に関するクイック回答

NOA

Is NOA any good? How to evaluate?

  • Trust long-window multi-city data, not one-offs
  • ≥1,000 km/model, ≥3 cities, day/night/peak under one standard
  • Delivery: method notes, sample disclosure, replayable evidence

How to compare urban/highway coverage?

  • Align samples first
  • Slice by city/road/weather; unify windows and mileage
  • Trends/heatmaps with sample sizes and windows

How many overrides are safe?

  • Use hazardous km/override, not raw counts
  • Scene-specific thresholds; filter prompt-only overrides
  • Show distributions and percentiles for risk alignment

How do you count hazardous overrides?

  • Automated + human, fully replayable
  • Auto-tag hard brake/steer/forced override; align video/trajectory
  • Human-review boundaries; frame-level replay/export

Compliance

How to test 2025 L2 compliance?

  • Checklist, item by item
  • Signal-level DMS/ADAS + behavior-level video checks
  • Output: per-item pass rates + non-compliant clip list

How to validate DMS effectiveness?

  • Align gaze/fatigue features with video
  • Quantify TPR/FPR; break down by cohort/time/env
  • Provide threshold guidance

Testing & Validation

How to validate ADAS quickly?

  • Templated, auto stats, closed loop
  • Standard templates + detectors; auto aggregate pass/fail and causes
  • Integrate ticketing (perception/planning/control)

How to handle long-mileage cost?

  • Crowdsourcing + automation reduces cost
  • Plan by sample×city×time
  • Use km-scale samples for stable estimates

Is crowdsourced public-road testing compliant?

  • Yes, with controls
  • Signed drivers, RBAC, full audit trail
  • Anonymize/minimize before disclosure

How to review corner cases?

  • Find common failure traits, make a replay set
  • Scene library + video slicing
  • Identify miss/double-detect/rule conflicts

Can we replay issues during tests?

  • Fully replayable
  • Sync interior/exterior video, sensors, event timeline
  • Frame-level RCA and exportable evidence

Competitors & Versions

Can we compare competitor NOA directly?

  • Yes, with highlights/lowlights
  • Cross brand/model/version, radar/compare tables, evidential clips
  • Disclose sample sizes and windows

How to accept vendor AD stacks?

  • Checklist-based acceptance and tiered grading
  • Multi-vehicle synchronous capture
  • Produce remediation items and re-test plan

How to compare versions of the same model?

  • Align samples per model/version
  • Compare NOA/ADAS core KPIs
  • Quantify iteration deltas and regression list

Experience & Marketing

How to extract NOA selling points for a launch?

  • Auto-generate “sellable” KPI packs
  • Leaderboards and visuals; one-click export
  • Include method notes and samples for release

Can evaluation assets be used in marketing?

  • Yes
  • One-click export charts/video for long/short content
  • Include sample/time/region scopes for consistency

Why do user reviews differ from technical KPIs?

  • Comfort metrics explain perception gaps
  • Use longitudinal/lateral accel and following variance
  • Provide tuning guidance

Insurance

How does NOA data support UBI pricing?

  • Risk features can go into models
  • Use hazardous km/override, hard brake/turn
  • Segment risks by scene for better pricing

How do insurers distinguish low-risk drivers?

  • Build risk personas by scene/behavior
  • Pair with incentives (points/cashback)
  • Lower loss ratios and adverse selection

Product & Download

How to download and install the CHEK app?

  • On the Products page tap Android or scan the QR
  • iOS/IVI will be announced upon release

How is data privacy protected?

  • Data minimization, anonymization/minimization before disclosure
  • RBAC and full audit trail
  • Itemized compliance acceptance

How to book a platform demo?

  • Use “Book a demo” on Products page
  • Or contact via footer phone

How to become a partner / join the Open Alliance?

  • Click “Join now” in Open Alliance module
  • See Open APIs links for integration

How to contact support?

  • Phone: (86) 18410870301
  • Or leave a note via demo/partner forms

Methodology & Scope

Stats window

  • Default rolling 90 days; split when major versions change

Sample representativeness

  • City slices by time/weather; standardized quotas for cross-city

Hazardous override definition

  • Overrides with safety risk; prompt-only excluded

Denoising

  • Drop device errors, missing video, interference; keep audit logs

Auditability

  • Traceable to window/route/clip IDs; supports third-party witness

Glossary

MPI (Miles per Intervention)

  • Km per override; higher is more stable

Hazardous km per override

  • Average km/override with safety relevance

Coverage/availability

  • Time/km ratio NOA works under given conditions

Pass rate

  • Completion rate per defined scene action

Comfort metrics

  • Longitudinal/lateral accel distribution, following variance