UX case studyVR prototypingUniversität Siegen
Who does what?
At a road accident, the hard part is rarely the bandage. It’s that nobody leads, nobody calls, and everybody waits for someone else to start. This is how we designed a VR training for exactly that moment.
Act IHow we arrived at the problem
VR already works for first aid. For one person at a time.
Immersive VR improves procedural knowledge, confidence and transfer. That isn’t the gap. The gap is that every system trains an individual: CPR quality, bandaging, triage, alone.
Mapped on two axes, procedural vs collaborative and professional vs lay, existing systems fill three cells. The fourth, collaborative and lay, is empty. It’s exactly where most real road accidents happen.
“Social and multi-user aspects for collaborative VTEs have received little investigation so far.”CHI 2022 workshop on collaborative virtual training
01The bystander problem
Several people arrive. Nobody moves.
Road accidents rarely involve one helper. Bystanders arrive within minutes, well inside the 8–12 minutes before an ambulance, and what happens next depends on coordination: someone calling, someone securing, someone with the injured.
Courses train the helper as a solo agent. With several people present and no coordination, help is delayed or never given: diffusion of responsibility. Everyone assumes someone else will act.
So the question isn’t what to do.
It’s who does what.
Act IIHow we approached it
Two decisions before designing anything.
I conducted one of the seven interviews myself. The team shared one interview guide, so every expert answered the same core questions.
Act IIIWhat we asked, and what it changed
Why it mattered
02Who we talked to
Seven people who have stood at real scenes.
Paramedics, an emergency physician, a fire service EMT, a first-aid training head, a disaster relief trainer, and one community responder with lived experience rather than clinical training.
Purposive sampling, at least two years of field or teaching experience, 30–60 minute semi-structured interviews, recorded with consent. Select a figure to meet them. Codes and roles only, no personal details.
03How we analysed it
Seven transcripts, five patterns.
We coded every transcript with Grounded Theory: open, then axial, then selective coding, so categories emerged from the data instead of being imposed on it.
Saturation was reached across all seven interviews: no new categories were emerging by the last transcripts.
Act IVFrom findings to concept
Every design decision traces back to something we heard.
04Three collaboration mechanics
Roles that emerge instead of being assigned.
- How it works
- Why
- Hypothesis
05Scenario design
Realism is a dial, set per tier.
One to three casualties, two to three players. Each tier adds casualties and takes guidance away, so a group is allowed to succeed before anything gets harder.
06One run, four phases
BuiltSecure the scene
The rescue chain as a scaffold: a phase can’t be skipped because the next one looks more urgent. Dense particles are built; dim, drifting ones are concept.
Act VHonest status
The concept, and what exists today.
- Runs in the browser, on desktop and VR headset
- Easy tier through the rescue chain, in order
- Solo play, or a team of AI teammates with role switching
- Asymmetric views: hazards for Scene, vitals for Medical
- Co-located multiplayer on standalone headsets (Unity)
- Explicit Delegation and confirmed handover between people
- Medium and Extreme tiers
- Bird’s-eye replay debrief
07Limitations and next steps
Justified by research. Not yet measured.
Next is a user study testing H1–H3 against baselines, then longitudinal work on retention past the 2–6 month decay window, the very problem this project set out to address.
08Reflection
The competence gap isn’t knowledge. It’s dividing the work under stress.
Teamwork under pressure is a trainable skill, like CPR. So instead of leaving a group to figure out coordination on the day, the training makes coordination itself the exercise.