100%
Reduced phone reliance (n=15 scenario responses)
Portfolio
MSc thesis · Kingston University London · Apr 2026
GIVE THE VEHICLE A MIND
An InCar HMI concept that detects when delay becomes meaningful and surfaces the next best action, instead of waiting for the driver to ask. Built to test whether AI in the car belongs in a conversation window — or in the workflow itself.
93%
Perceived helpfulness
100%
Trust under bounded autonomy
02 -
The problem
The problem
DELAY IS NOT A TRAFFIC ISSUE. IT'S A COORDINATION PROBLEM.
When delay affects a meeting, reservation, or arrival, drivers fall back to their phone. They call, message, check maps — while in motion.
There is no proactivity by the system itself.
NHTSA, 2023
3,275
Deaths from distracted driving, US, 2023
Current in-car systems don't help with the consequence. The phone does. The phone also kills.
03 -
The question
The question
HOW CAN AI IN THE CAR SUPPORT DRIVERS WHEN DELAY BECOMES MEANINGFUL — WITHOUT BECOMING ANOTHER SOURCE OF DISTRACTION?
04 -
Research
Research
FIVE DRIVERS. SAME PATTERN.
5
Semi-structured driver interviews
5/5
Reported delay-related stress
4/5
Described phone fallback explicitly
Delay was consistently described not as a navigation problem but as a coordination one. When ETA shifted past a meeting, reservation, or arrival, drivers reached for the phone.
My brain is packed up — I'd want help and suggestions from the vehicle.
05 -
Design decisions
Design decisions
FOUR CHECKPOINTS.
01
Narrowed to delay
Cut the broader assistant concept. Focused on the moment delay becomes consequential.
02
Workflow-native, not chatbot
Designed support to live inside the existing HMI surface, not as a separate AI mode.
03
Bounded autonomy
Suggested actions, never executed silently. Confirmation required for messages, joins, reschedules.
04
Context fusion as relevance filter
Used calendar, ETA, and route together to decide when support is meaningful, not just if.
06 -
The prototype
The prototype
FIVE SCENARIOS. STATE-DRIVEN SIMULATION.
Built as a state-driven simulation rather than a click-through. Each scenario shows how the interface responds when route, ETA, and commitment data converge.
Scenario set
| Scenario | Type |
|---|---|
| Late for Meeting | Core delay |
| Late for Reservation | Core delay |
| Running Home Late | Core delay |
| Tiredness | Exploratory |
| Cabin Discomfort | Exploratory |
Three core delay scenarios plus two exploratory vehicle-state scenarios.
07 -
Validation
Validation
FOUR HYPOTHESES. TESTED WITH THE SAME FIVE DRIVERS.
Each hypothesis was assessed across the three core delay scenarios (Meeting, Reservation, Late Home) — 15 total scenario responses.
H1Supported
Phone reliance reduced
0%15 / 15
Every participant said the support would replace reaching for the phone.
H2Supported
Proactive support felt more helpful than current systems
0%14 / 15
14 of 15 scenario responses positive.
H3Supported
Trust held under bounded autonomy
0%15 / 15
Participants accepted suggestions. Rejected silent execution.
H4Directional
Distraction reduced
0%12 / 15
Held in principle. Presentation layer still needs tightening — wording, hierarchy, option count.
Hypothesis results from 5 participants × 3 core delay scenarios.
Three of four hypotheses validated strongly. The fourth held in principle but exposed iteration points.
08 -
Technical
Technical
BUILT AS A SYSTEM, NOT A SCREEN SEQUENCE.
JSON scenario data drives timing, triggers, and content. A JavaScript scenario engine sequences states. React components render the interface dynamically.
The architecture mirrors how a real vehicle context engine would have to behave — not a static screen sequence.
09 -
Next steps
Next steps
WHAT THIS PROJECT IS MISSING.
- 01Tighten the three validated scenarios into the product core.
- 02Refine the distraction layer — hierarchy, wording, option count.
- 03Move from perceived-distraction self-reports to glance behaviour and timed-task measures.
- 04Explore the underlying architecture as a real context engine — live services, permission handling, confidence thresholds.
000 · Back to portfolio
Moritz Wallbrecher→