Mixed Reality and Passthrough: When Safety Training Leaves the Closed Headset

A maintenance technician puts on a headset in the workshop. In front of him is the real electrical panel, with its own cables and labels; above the disconnector a digital overlay shows the lockout/tagout sequence, and the system will not let him proceed until the lock is placed in the right spot. This is not virtual reality: it is mixed reality in passthrough. And it changes how training is designed.

Three different technologies, not three names for one thing

The most widely used conceptual reference is still the reality-virtuality continuum proposed by Milgram and Kishino in 1994, echoed by ISO/IEC standardisation work on reference models for mixed and augmented reality. Along that axis sit technologies that answer very different training needs.

Immersive virtual reality

The headset is closed and replaces the environment entirely. The learner is inside a fully synthetic scenario: a fragile roof, a confined space, a smoke-filled department. It is the only option when the scenario is impossible, costly or dangerous to reproduce for real. We covered it in VR and 360 video for safety training.

Augmented reality

Digital content overlaid on the real world, usually without the system reconstructing the geometry of the space. It runs on smartphones, tablets or smart glasses and anchors content to markers, QR codes or flat surfaces. It is the logic behind many assisted maintenance applications.

Mixed reality

Here the system maps the real space (mesh, planes, occlusion) and virtual objects behave accordingly: they disappear behind a pillar, rest on the workbench, are covered by the operator's hand. MR can be delivered through transparent optics (optical see-through) or through video passthrough: the headset's outward cameras capture the environment and re-present it on the internal displays.

Passthrough, then, is not a fourth category: it is the technical mode by which most headsets on the market today deliver MR experiences while reusing VR hardware.

What the research says

Systematic reviews published in Safety Science offer a cautious but encouraging picture. The 2023 meta-analysis on VR for safety training reports that immersive VR is the most studied modality and the one with the most solid results in learning and engagement. The subsequent 2024 systematic review on AR indicates that augmented reality entered this field later, but with positive outcomes.

Particularly useful is a 2024 study, also in Safety Science, that directly compared fire safety training in video see-through AR (that is, passthrough) with VR and with conventional video. Immediately after training, AR outperformed traditional video in knowledge retention, long-term self-efficacy and perceived quality of instructions. It is one of the few direct comparisons available across the three modalities.

A recurring point emerges from this literature: the advantage does not come from the device, but from letting people practise risky decisions under safe conditions.

When to use what

Choose immersive VR when:

• the real scenario is dangerous, unrepeatable or inaccessible (fire, evacuation, work at height, confined spaces);
• you need total control over variables and unexpected events;
• you want to deliver the identical exercise to dozens of people across sites.

Choose passthrough MR when:

• the competence is tied to a specific plant or piece of equipment the learner is standing in front of;
• you want to drill the procedural sequence on the real workplace, as with lockout/tagout on machinery;
• the session is in pairs or teams and people need to see each other's faces;
• you need to reduce the isolation anxiety of first-time headset users.

Choose tablet or smart-glasses AR when:

• the goal is task support more than simulation;
• the scale is wide and the hardware budget limited.

Designing a mixed reality module

A few practical rules separate a demo from a real training intervention.

• Start from the task, not from the 3D content. Define the sequence of critical decisions, then decide what must be virtual.
• Use the virtual only for what cannot be real. If the electrical panel is there, do not model it: add residual voltage, arc flash, warnings.
• Design the physical space. You need a defined clear area, free of obstacles, with a second person acting as spotter.
• Keep sessions short. Six to twelve minutes per scenario, with the option to repeat.
• Debriefing is part of the module, not an appendix: without reflection the experience stays a memory, not learning.

Limits and cautions

Cybersickness. A systematic review published in 2025 across twenty-five studies (2013-2024) found that the most frequent effects are oculomotor disturbances and disorientation, and that in roughly half the studies scores were low or inconsistent. The main cause remains sensory mismatch between perceived and actual motion. Passthrough reduces part of the problem because the real horizon stays visible, but it introduces camera latency and distortion: test it, do not assume it.

Hidden content maintenance costs. The real cost is not the headset, it is keeping scenarios current when plants, PPE or procedures change. Budget an annual content review and name an internal owner.

Hygiene, comfort and accessibility. Washable face interfaces, vision correction, handling of people who wear glasses, an alternative path for those who cannot tolerate a headset: without this, participation drops.

It does not replace mandatory hands-on training. MR prepares, drills and verifies; it does not remove the field practice required by regulation and company procedures.

How to measure

Avoid the enthusiasm of "they loved it". Collect comparable indicators:

• critical errors per scenario and time to first correct action;
• delayed retention (30-90 days), not just an immediate test;
• perceived self-efficacy before and after;
• observed behaviours on site and reported near misses;
• cost per completed learner, compared with classroom delivery.

A coherent measurement setup is described in KPIs and ROI of safety training.

In short

The right question is not "VR or MR?" but "how much of the real world does this learning objective need?". If the real context matters, passthrough is today the most practical route. If the context is impossible, closed VR remains. And if the goal is remote collaboration, there are equally effective non-immersive formats, such as digital escape rooms for distributed teams.