Back to work
Automotive · HMI

Hyundai IONIQ 6 — designing for trust at L2+

A sponsored HMI project exploring driver confidence in Hyundai’s HDA 2.0 system. Real-driving research, simulator testing and three interface directions shaped an adaptive instrument cluster for the IONIQ 6.

Role
HMI Designer
Period
Jan — May 2025
Client
Hyundai · HATCI Lab
Themes
HMI design · Driving research · Interactive prototyping
Chapters
Research

Clarity at the handoff

How might we help drivers feel more confident when control shifts between the human and Hyundai’s HDA 2.0 system? This sponsored HMI project explored the IONIQ 6 experience in Level 2 and Level 2+ assistance scenarios, focusing on system status, information hierarchy and feedback at critical transitions. The stakeholder map placed Hyundai and the HMI lab alongside drivers, other road users, regulators and groups with different accessibility needs. It helped frame the interface as part of a wider driving experience, with expectations that extend beyond the person looking at the cluster.

Original IONIQ 6 cluster

HDA Auto Disabled by

  • Traffic JamOnly smart cruise on
  • AcceleratingFlashing White
  • Collision AvoidanceOff temporarily
  • High influence , Low importance

    • Regulatory Bodies
  • High influence , between low and high importance

    • Wayne Li
  • High influence , High importance

    • Hyundai Labs
    • HMI Lab
  • Low influence , Low importance

    • Pressure Groups
  • Low influence , High importance

    • Hyundai Drivers
    • Elderly People
    • People with Disabilities
    • Other Drivers on Road
    • Hyundai Competitors

Observe driving in two environments

Drivers completed tasks in both a real vehicle and a custom Unreal Engine simulator. The setup combined ECG-based heart-rate variability, Tobii eye tracking and self-report surveys. A steering rig, highway rendering and surround sound allowed the team to explore assistance behavior in a repeatable setting before comparing it with the real-driving experience.

Real vehicle

A real-driving session recorded from the back seat
A participant driving the IONIQ 6, an observer beside them
A participant wearing Tobii eye-tracking glasses

Unreal Engine simulator

A participant at the simulator
The simulator wheel and its HDA button
The simulator rig

Simulator rig

  • Buttons that correspond to semi‑autonomous functionality
  • Centering force tension in corrective modes
  • 2Environ­ments
  • 8Tasks
  • 9Points of data collection

Combine workload, behavior and feedback

The Driving Activity Load Index captured attention, visual and auditory demand, time pressure, interference and situational stress. We considered those responses alongside task observations and biometric data, then organized the feedback into an affinity map. Exploratory comparisons between simulator and car measures informed the next study round.

Situational stress was weighted highest but ranked fifth by raw score; auditory demand was weighted lowest but ranked second.

How DALI works

Survey: Driving Activity Load Index

NASA TLX but for driving

Part 1

Effort of AttentionHow hard did you have to work to accomplish your level of attention?

Visual DemandHow visual effort was required for the task?

Auditory DemandHow much listening effort was required for the task?

Temporal DemandHow hurried or rush was the pace of the task?

InterferenceHow much did the simultaneous execution of supplementary tasks disturb your driving performance?

Situational StressTo what degree did you experience environmental constraints or stress while driving?

Part 2 · 15 pairs

  1. 1Effort of Attention
  2. 2Visual Demand
  3. 3Auditory Demand
  4. 4Temporal Demand
  5. 5Interference
  6. 6Situational Stress

Please circle the dimension that had a greater impact on your driving experience

Correlation (LKA Task)

Overall correlation between the simulator (s) and the car (c)

Exploratory
  • Time To Performance (TTP)rTTP
  • Time To Recognition (TTR)rTTR
  • Time To Task (TTT) = TTR + TTPrTTT
  • ECG (as a function of time)rECG
  • Heart Rate (as a function of time)rHR

R=rTTP+rTTR+rTTT+rECG+rHR5

Design

Turn research into interface requirements

Round 1 surfaced three connected problems: ambiguous activation and status cues, a cluttered visual hierarchy, and too little timely reassurance about what the assistance system was doing. Together, these made handoff moments harder to interpret. The design opportunity was to communicate the right state and the right information at the moment it mattered. The first requirements were concrete: make entry into and exit from HDA visible, distinguish active and interrupted assistance, and keep important feedback readable without obscuring the driving view. These requirements connected each pain point to a behavior the interface could communicate.

Breakdown in System Communication

Erodes Trust

Automation should be clear, predictable, and reassuring to help users feel confident and reduce confusion.

I actually don’t know this what the system’s supposed to do. And I’m like, so scared”

From the affinity map

User Need: Clear Change to HDA

  • Page Change Format for When HDA is Active
HDA ON
Turned grey

Cognitive Overload from Poor UI Hierarchy

Increases Mental Effort

The UI should highlight key information at the right time, working like a helpful assistant instead of overwhelming users

I’m still confused with the steering wheel and the lines on both sides of the lane like what the difference of those two are”

From the affinity map

User Need: Strong Indication

  • Keep Information
  • Alerts, notifications, and visual cues
  • Clear Prioritized Layout
Accumulated Info

Lack of Engagement & Reassurance in Autonomy

Creates a Trust Gap

Effective automation keeps users engaged with clear, timely feedback that builds trust without overwhelming them.

I don’t know. I feel like I sometimes don’t trust my car’s cruise control”

From the affinity map

User Need: High Engagement

  • balance between control and delegation
  • Without interference
Info Table Block HDA Animation

Explore the hierarchy before the visual finish

Low-fidelity prototypes separated driving status from secondary information and explored layouts for manual driving, active assistance, lane detection and heavy traffic. Comparing these states side by side made it easier to decide what should stay stable and what should change with the driving mode. The exploration expanded into a family of instrument-cluster states. Layout, vehicle imagery, status treatments and feedback were refined together, so the proposal could communicate changes in assistance rather than only present a polished static view.

Lo-Fi Prototypes

  • Lane Keeping(on its own)
  • Lane Following
  • Following Distance
  • Smart Cruise Control+ Lane Following Turned on Condition+ Following Distance Turned on Condition
  • HDA+ Smart Cruise Control Condition

+ Smart Cruise Control Condition

ActivePausedDisengaged
Center
Center
Distance 4
Distance 4
(Glowing Effect)
(No Glowing Effect)

Test three directions

We developed three interface directions for comparison. White UI established a minimal baseline. Green UI emphasized reassurance and clear assistance feedback. Blue UI explored a more technical, information-focused presentation. The study compared these directions through driving tasks and participant feedback.

What differsGrey icons, white lane lines, and wedges for speed and range

Validation

Round 2: compare, co-design and refine

The second round combined visual UI testing with functional testing in the simulator and car. Whiteboard co-design let participants rearrange information and discuss what they needed in each driving mode. Their feedback pointed toward an adaptive hierarchy: speed mattered more during manual driving, while the vehicle visualization became more important with assistance engaged.

Study plan

Visual UI Testing

Phase 1:Baseline Testing
Phase 2:Rapid Iteration Co-Design

Functional Simulator/Car Testing

Phase 1:Car vs Sim + original UI
Phase 2:Sim only + UI prototypes

Criterion

Confidence Metric measurements must achieve at least a 10% relative increase from the original UI measurement to demonstrate a meaningful and practically significant improvement in user confidence when evaluating our UI designs.

Process

  1. 1Co-Design Rapid Iteration in FigJam with test subjects
  2. 2Identify insights, pain points, and trends based on step 1
  3. 3UI North Star iteration based on phase 2

Co-design tokens

  1. Speedometer
  2. Icon Cluster - Vehicle Status
  3. Cruise Speedometer
  4. Vehicle Illustration
  5. Alerts/Nudges
Sample Screen

Level-2, the co-design Sample Screen: Alerts/Nudges on the left with the Cruise Speedometer under it, the Vehicle Illustration in the large slot with the Icon Cluster - Vehicle Status under it; the Speedometer is not on this screen.

Findings

  1. In semi-level 2 scenarios, LESS info = greater confidence. In fully level‑2 scenarios, MORE info = greater confidence

  2. Users wanted there to be a clear differentiation in information size and placement when switching between manual driving and fully level‑2 driving scenarios

    (Speedometer, Icon Cluster - Vehicle Status, Cruise Speedometer, Vehicle Illustration, Alerts/Nudges)
  3. Components such as speed play a more important role in manual driving while the vehicle illustration plays a bigger role in level‑2 scenarios

    (Speedometer, Vehicle Illustration)

Make confidence measurable

The study used a custom confidence score to bring physiological signals, task timing and self-report data into a common framework. The documentation records how data was prepared, aligned and weighted. The pipeline also separates implemented inputs from planned facial-emotion and eye-tracking extensions, keeping the measurement model open to further development. Paired-sample analysis compared simulator and real-driving measures from the same participants. The analysis considered confidence scores, UX self-reports, pupil size, heart-rate variability and task time, giving the team several perspectives on the experience across environments.

Confidence Metric Formula

C=100×(1−(α·HRVHRVmax+β·HRdiffHRmax+γ·ECGvarECGmax+δ·T))

α, β, γ, δ = Weighting coefficients

Data pipeline

Implemented: ECG.csv goes through ecg.py to HRV/HRVmax and ECGvar/ECG; BPM.csv goes through bpm.py to HRdiff/HRmax and T; Google Form gives SS. These feed confidence.py, which outputs the confidence metric. Planned: Cam Data through FER software to FS, which also feeds confidence.py; Tobii Data through tobii.py to LS, not yet connected.

Paired-Sample t-test Process: Simulator vs. Real Driving

Driving SimulatorReal Driving Condition
  • Confidence score, measured in both environments. Source: Custom confidence metric. Focus: Reported certainty and comfort.
  • UX self-report, measured in both environments. Source: Custom UI questionnaire. Focus: Usability, clarity and satisfaction.
  • Pupil size, measured in both environments. Source: Tobii Pro Glasses 2. Focus: A signal considered in cognitive-load analysis.
  • HRV (RMSSD), measured in both environments. Source: Polar Beat ECG. Focus: Physiological variation during driving tasks.
  • Task time, measured in both environments. Source: Simulator and real-car interaction logs. Focus: Time and friction in completing tasks.
What each step does
  1. 1Same participants test both environments
  2. 2Measure performance in both simulator and real driving
  3. 3Subtract simulator measurements from real driving values
  4. 4Average differences divided by standard error of differences
  5. 5Check p‑value to determine if differences matter

Assess HMI Design Effectiveness Across Environments

Green UI led the prototype comparison

The project study reported that Green UI achieved the highest confidence score, with a 10% improvement over the original UI. The comparison chart shows how the three directions performed for the participants shown. This result guided the final prototype toward clearer assistance states and information that adapts to the driving mode.

Confidence Metric Results

  • White UI
  • Blue UI
  • Green UI

P011

P011: Green UI 36.66, Blue UI 35.35, White UI 22.1

Green UI

LKA
56.5
LFA
14.44
SCC
39.03

(56.5 + 14.44 + 39.03) / 3 = 36.66

Blue UI

LKA
47.7
LFA
not recorded
SCC
23

(47.7 + 23) / 2 = 35.35

White UI

LKA
44.2
LFA
not recorded
SCC
0

(44.2 + 0) / 2 = 22.1

Each total equals the mean of that direction’s task scores on the calculation board.

Participants shown: P011, P013, P007, P014

Final concept

A clear system of assistance feedback

The final Green UI brings together a compact status panel, a central vehicle illustration and a dedicated steering-assistance indicator. Dynamic icons distinguish paused and active states, while lane-keeping and following-distance feedback keep the assistance behavior visible. Information density changes with the driving mode. A shared system brought typography, icons, colors, vehicle illustrations and assistance components into one vocabulary. The component work focused on making the difference between normal operation, active support and a request for driver attention easier to recognize. Three prototype sequences show the Green UI in motion, extending the component system into changes of state and feedback over time.

Green UI · rebuilt in code

Color

  • Active support
  • Normal operation
  • Request for driver attention

Prototype videos

North Star: bring feedback closer to the driver

The North Star concept explored a steering-wheel-mounted display and a fixed central screen within the turning wheel. Sketches and screen studies considered shorter eye travel and more immediate feedback, alongside the challenges of motion, rotating content and cognitive load. This remained a concept for further exploration. The cabin visualizations bring the steering-wheel concept and main display into a wider interior experience. Next steps were to improve simulator–UI integration, create opportunities to learn the assistance system through interaction, and add a more complex cityscape. Further testing with drivers familiar with Hyundai vehicles would help examine the wide variation in feedback observed in this study.

Screen and Airbag stay fixed

Wheel-mounted screen

Pros

  1. Reduced Eye Travel Distance
  2. Natural Eye-Hand Coordination
  3. Consistent Visibility Regardless of Steering Position

Cons

  1. Motion artifacts from turning the wheel
  2. Cognitive Load with rotating content
  3. Lacks research, though a 2021 study by Igarashi et al. examined steering wheel–mounted touchscreens in EVs, noting increased driver engagement
North Star cabin visualization with a compact display in the steering wheel.
North Star cabin visualization with a compact display in the steering wheel.