What does automotive-hmi do?
Every glance at a screen is a glance away from the road. Automotive HMI is the most quantified high-stakes domain: regulators, rating bodies and platform owners all publish numbers. This skill turns them into design rules, and is honest where evidence is missing (two-wheelers, EV, HUD).
This skill stands on its own. For alerts, automation trust and payments in the car, the companion criticalUX skills go deeper.
How to use this skill
- Classify the surface and its driving state. Cluster, centre display, HUD, steering-wheel controls, voice, phone projection (CarPlay/Android Auto), native app (Android Automotive), two-wheeler TFT, companion phone app. Then: parked, driving, or under automation (which SAE level).
- Give every task a glance budget before designing it (
distraction-and-glance.md). If a task cannot fit, move it to voice, a physical control or "parked only". - Map controls to physical, touch, voice or automatic using the control map.
- If any automation is involved, design mode display, escalation and handover with
automation-and-handover.md. - Design the visual layer with
clusters-hud-and-visual.md. - For bikes and scooters, load
two-wheelers.mdand label borrowed car thresholds as borrowed. - Test with occlusion or eye tracking against the budgets; compare to a physical-button baseline.
| Reference | Load it for |
|---|---|
references/distraction-and-glance.md | NHTSA glance and occlusion budgets, lockouts, display angles, Euro NCAP physical controls, Android for Cars and CarPlay limits, benchmark test data |
references/automation-and-handover.md | SAE levels, R171/R157, IIHS safeguards, driver monitoring, mode display, escalation, takeover times, Tesla recall lessons |
references/clusters-hud-and-visual.md | Cluster layout, tell-tale colours, speed and limit display, typography, night palettes, HUD, zero-layer surfacing, haptics |
references/assistants-ev-ota.md | Voice and LLM assistants, EV range/charging/regen, OTA updates, connected-car phone apps |
references/two-wheelers.md | Motorcycle and scooter dashboards, glove input, navigation, safety features |
references/case-studies.md | What OEMs did and what it teaches |
The rules that matter most
Source types: [REG] regulation · [STD] standard, rating protocol or platform guideline · [RES] published research · [PRAC] company practice · [INF] our synthesis from the sources.
- Glances ≤ 2 s, task total ≤ 12 s. A task must also be completable in 1.5 s occluded glances within 12 s total [STD] (NHTSA 2013, voluntary).
- One hand, interruptible at any moment, resumable where the driver left off [STD].
- Lock while moving (propulsion on and not in Park): video, auto-scrolling text, manual text entry for messages or browsing, reading web, social or message text. Read aloud instead [STD].
- Display centre within 30° below the driver's line of sight [STD]. Measured cars needed 20°–56° to reach the bottom of their screens [PRAC].
- Physical controls for indicators, wipers and washer, hazards, horn and the SOS/eCall button (Euro NCAP 2026). Keep climate and volume physical where possible [STD/PRAC].
- Frequent tasks in ≤ 3 taps; ≤ 5 screens per task; driving text 1–3 lines and ≤ 120 characters (Android for Cars) [STD].
- Benchmark against buttons. A 2005 Volvo with buttons finished four tasks in 10.0 s; touchscreen cars took 13.7–44.9 s at 110 km/h [PRAC, magazine test].
- Automation mode is unmissable and simple to engage and disengage. Weak prominence led to a 2-million-vehicle recall (Tesla 23V-838) [REG].
- Monitor eyes and head, not just hands on the wheel (UN R171) and escalate: visual → audible → haptic → slow to a stop with hazards → lock the feature out for the drive (IIHS) [REG/STD].
- Plan handovers for the slow driver, not the average one. Takeover times ranged 1.9–25.7 s; give at least 10 s for Level 3 and a minimal-risk manoeuvre if nobody takes over [RES/REG].
- Never blank a safety value. Show the last known speed limit and say plainly when a sign couldn't be read [STD].
- Key functions at the top level (zero layer); suggestions accepted or dismissed in one tap [PRAC].
- One typeface, restricted palette, colour reserved for status (tell-tales); cluster and centre display agree [PRAC/INF].
- Two-wheelers: one glove-friendly input, minimal glance mode by default, auto day/night, and say which car thresholds you borrowed [PRAC/INF].
Before you hand over, check
- Every task listed with its driving state and glance budget; over-budget tasks moved to voice, physical or parked-only
- Lockouts implemented and explained ("Available when parked"), with voice alternatives
- The five Euro NCAP functions on physical inputs; climate and volume reachable without menus
- Menu depth ≤ 3 taps for frequent tasks; text ≤ 120 characters while driving
- Automation states (off, available, active, degraded, takeover requested, minimal-risk manoeuvre) each designed, in the cluster and HUD
- Escalation sequence with timings; lockout after repeated misuse
- Speed limit shows last known value and an "unreadable" state
- Night palette tested for glare; text sizes tested at the real viewing distance
- Tested with occlusion or eye tracking, and against a physical-control baseline
- For two-wheelers: glove input, sunlight legibility, vibration, phone-off behaviour, signal loss
Standards behind several rules (ISO 15005, 15008, 16951, 2575, SAE J2364, J3016, JAMA) are paywalled and were not verified; where this skill cites them it says so.