GIVE THE VEHICLE A MIND

INTERACTIVE DEMO / FINAL EXPERIENCE

The quickest way to understand the concept is to watch the journey change meaning. The interface starts as a normal drive, detects that delay now threatens a commitment, offers bounded support, then returns to the driving state after confirmation.

Scenario set

Nothing sensitive happens silently. The car prepares the message, offers the remote join, and drafts the location share — the driver confirms with one tap or voice. — Bounded autonomy line

THE INSIGHT / WHY THIS MATTERS

Delay is not experienced as a traffic problem. It becomes a coordination problem the moment ETA collides with a meeting, reservation, or person waiting at home.

Most in-car AI assistants still behave reactively: the driver asks, the system answers. That misses the high-stress moment when the driver is already managing time pressure and is most likely to reach for a phone.

That fallback sits inside a real safety problem. NHTSA reports 3,275 deaths in crashes involving distracted drivers in 2023.

Interview evidence · n=5

"

Trapped in the car.

"

Called a coworker to stall for time before a meeting.

"

My brain is packed up.

"

There is no proactivity by the system itself.

"

Not very useful.

How might an in-car AI system use contextual awareness to support drivers during delay-related, time-critical journeys without increasing distraction or breaking mental flow? — Problem statement

The concept response was deliberately narrow: not another assistant, and not a chat layer. A vehicle should notice when a routine journey has become consequential and surface the single next best action in the existing HMI.

Core proposition: AI as embedded vehicle behaviour, not as an added tool.

PROCESS & RESEARCH

Secondary research set the safety constraints: short-glance design, limited eyes-off-road time, and human-AI interaction principles from Amershi, Shneiderman, and Norman. Primary research made the product logic sharper: delay only matters when it breaks a commitment.

I used the research to make interaction decisions rather than to document a chronological design process.

Artefacts produced+

• 012 proto-personas: time-pressured business professional and socially-driven younger user.
• 02User journey map mood-coded across 6 stages from before trip to after support.
• 03Mental model sketch: current manual delay management vs. proposed next-best-action support.
• 04Information architecture: Input → Interpretation → Output with progressive disclosure.

VALIDATION / EVIDENCE

I ran follow-up sessions with the same 5 participants and tested the concept against four hypotheses. Percentages are positive responses out of 15: 5 participants across 3 core delay scenarios.

Where participants drew the line

Participants did not reject proactivity. They rejected irrelevant or overreaching proactivity. The same prompt at the wrong moment would have failed; in the core delay scenarios, the system appeared only after delay had clearly broken a commitment.

Evaluation artefacts and backlog+

• 01Stronger confirmation feedback after an action: sent or shared state.
• 02Tighter visual hierarchy on the action card with a clearer primary action.
• 03Reduce option count in some scenarios.
• 04Add Share Location as a first-class option in Late Home.
• 05Justify the value of exploratory scenarios more clearly, or drop them.

TECHNICAL / SYSTEM THINKING

The challenge was not generating AI responses. It was deciding when context was meaningful enough to interrupt the HMI at all, and then keeping the intervention bounded.

Technology logic and blueprint+

The system fuses signals that are not new in isolation, but meaningful together: route and live ETA, calendar timing, traffic, communication context, vehicle state, trip context, and learned behaviour patterns.

01

Delay detection

Has the ETA shifted meaningfully versus the planned arrival?

02

Commitment matching

Is there a calendar event, reservation, or known obligation tied to this trip?

03

Relevance assessment

Does the new ETA now break that commitment?

04

Situation classification

Work, social, family, or general context changes tone and action set.

A delay only triggers intervention when all checks line up. A slow drive with nothing scheduled stays silent. The same slow drive 20 minutes before a meeting surfaces support.

REFLECTION / NEXT STEPS

• 01The value lives at the intersection of route, ETA, and commitment.Without all three, the system has nothing meaningful to say.
• 02Trust does not equal automation.Trust came from the system holding back, not from doing more.
• 03Concept logic mattered more than the final pixels.The idea validated more cleanly than the presentation layer; distraction-related refinement is the next iteration.
• 04The evidence is still qualitative.Small sample, self-reported measures, and no glance-tracking or simulator-based workload testing.

Next step

Keep the next version focused on the three validated core scenarios: Late for Meeting, Late for Reservation, and Running Home Late. Refine hierarchy, confirmation feedback, and option count before claiming low-distraction performance.

After a larger sample and simulator testing with objective workload measures such as NASA-TLX and eyes-off-road time, the same framework can extend into adjacent areas such as fatigue and trip-mode personalisation.