AR vs VR: the difference, and which fits training and remote collaboration

What is the difference between AR and VR?

Microsoft's mixed reality documentation draws the line clearly. Experiences that overlay graphics, video streams or holograms on the physical world are augmented reality. Experiences that block your view to present a fully immersive digital world are virtual reality. The distinction is about your eyes, not the software: in AR you still see the real room, in VR you do not.

Two more terms sit around them:

• Mixed reality (MR) covers experiences that can move between AR and VR and place a digital object in the room as if it were physically there. The term comes from a 1994 paper by Paul Milgram and Fumio Kishino, which arranged displays along a "virtuality continuum" from the real environment at one end to a fully virtual one at the other.
• Extended reality (XR) is the umbrella term. The Khronos Group's OpenXR specification defines XR as a continuum of real-and-virtual combined environments generated by computers, inclusive of VR, AR and MR.

Vendors add their own names on top. Meta sells Quest 3 as a mixed reality headset, Google calls its headset and glasses platform Android XR, and Apple introduced Vision Pro in June 2023 as a "spatial computer". Under the labels, the comparison that matters looks like this:

How AR glasses and VR headsets show you the world

There are two ways to put virtual content in front of someone's eyes while they still see the room, and they explain most of the practical differences between devices.

Optical see-through. Glasses or a visor with transparent lenses let real light through, and a small projector or display adds light for the virtual content through an optical combiner. You see the real world directly, with no delay. HoloLens 2 works this way, and so do glasses with an in-lens display such as Meta Ray-Ban Display. The catch is contrast: the virtual image has to compete with daylight. A 2021 review in Light: Science & Applications calculates that to keep a 10:1 ambient contrast ratio on a sunny day, an AR display needs to deliver at least 30,000 nits, which is why see-through content often looks faint outdoors.

Video passthrough. A headset with opaque displays films the room with outward-facing cameras, adds the virtual objects and shows the combined picture on its internal screens. Meta Quest 3 and 3S and Apple Vision Pro work this way, which is how one headset can switch between mixed reality and full VR. Everything depends on how fast and how sharp the camera feed is: Apple says Vision Pro's R1 chip processes input from 12 cameras, five sensors and six microphones and streams new images to the displays within 12 milliseconds.

Both designs share one comfort problem. A headset shows 3D images on a display at a fixed focal distance, so your eyes converge on a virtual object at one distance while they focus at another. This vergence-accommodation conflict causes eye strain and fatigue in stereoscopic displays, and the same review notes it may be more serious in AR than in VR, because the virtual image sits directly on top of real objects at their real distances.

AR and VR headsets in 2026: what is available and what changed

The market in September 2026 looks very different from 2023. The general-purpose headsets now do both AR and VR through passthrough, Microsoft's see-through HoloLens is no longer made, and the next wave is glasses.

The changes worth knowing:

• Apple Vision Pro went on sale in the US on February 2, 2024. The M5 version, announced on October 15, 2025, adds a Dual Knit Band, up to 2.5 hours of general use per charge and up to 120 Hz. Per Apple's tech specs, it weighs 750 to 800 grams plus a separate 353-gram battery.
• Meta Quest 3 was announced on September 27, 2023, which Meta called the world's first mass-market mixed reality headset, with full-colour passthrough. Quest 3S followed on September 25, 2024 with the same mixed reality features at a lower price, when the 512 GB Quest 3 also dropped to $499.99. Both now cost more on the Meta Store, which rates their battery life at 2.2 hours (Quest 3) and 2.5 hours (Quest 3S).
• Android XR was announced on December 12, 2024 with Samsung and Qualcomm. Its first device, Samsung Galaxy XR, went on sale in October 2025; Google says it switches between full immersion and staying present in the real world, and ships XR versions of Google apps including Meet. At Google I/O in May 2026, Google said Android XR audio glasses with Gentle Monster and Warby Parker frames launch first, in fall 2026, with display glasses to follow, and reservations for XREAL AURA, wired XR glasses on Android XR, opened in June 2026.
• Meta's glasses. Meta showed Orion, its first true AR glasses, in September 2024 as a product prototype that will not reach consumers. What it sells is Meta Ray-Ban Display, whose display sits off to the side and is designed for short interactions, not all-day overlays. On September 23, 2026 it announced Meta VR Glasses: about 100 grams on the face, which Meta says is five times lighter than Quest 3, with compute and battery in a puck connected by an optical tether. What camera glasses record, and what the law allows, is covered in AI smart glasses and privacy.
• Microsoft stepped back. Its HoloLens release notes state that HoloLens devices have not been manufactured since December 2024, and that the November 2024 update was the final feature release for HoloLens 2, with monthly security updates through December 2027. On the PC side, Windows Mixed Reality is removed from Windows 11 version 24H2; existing headsets keep working with Steam on version 23H2 only until November 2026.

VR and AR for training: what the evidence says

The broadest summary is a 2021 meta-analysis in Human Factors that pooled transfer-of-training studies across fields. Training in VR, AR or MR produced the same outcomes as training in a conventional, non-simulated environment: equally effective, not better. The authors conclude that XR's value lies where conventional training is impossible, for example when danger or cost rules it out. That is the lens for every use case below.

Medicine and surgery

Surgery produced the first randomized, blinded test of whether VR skills carry over to real operations. In that study, published in Annals of Surgery in 2002, 16 surgical residents were split between VR simulator training and no simulator training. In real laparoscopic gallbladder operations, the VR-trained residents dissected 29% faster and were six times less likely to make errors (1.19 against 7.38 errors per case). The sample was small, but the errors were scored blind, on real surgery.

Across health professions, a 2019 meta-analysis of 31 randomized trials (2,407 participants) found that VR slightly improves knowledge scores compared with traditional learning (standardized mean difference 0.44) and improves cognitive skills more clearly (1.12, from 4 studies). The certainty of evidence was low to moderate, and none of the studies measured effects on patients.

AR shows a different profile. A 2025 meta-analysis of 29 publications found AR and MR teaching improved skill scores, lowered failure rates and shortened performance times compared with traditional teaching, but did not significantly improve knowledge, with the caveat that heterogeneity and publication bias may inflate the estimates. The pattern is useful: AR helps people do a procedure; it does not replace studying it.

Aviation

Aviation already trains in certified simulators, so the question is whether a headset-based device can qualify as one. In April 2021 the European Union Aviation Safety Agency granted the first certificate for a VR-based flight simulation training device, built by Loft Dynamics for rotorcraft pilots, so risky manoeuvres can be practised virtually; EASA notes that around 20% of rotorcraft accidents happen during training flights. A 2025 systematic review of XR in pilot simulator training found a positive, moderately strong effect on pilot performance (effect size 0.884), but only 5 of its 67 eligible studies qualified for the meta-analysis, and the authors flag simulator sickness and a lack of robust transfer trials as serious gaps.

Engineering and manufacturing

For manual assembly, a scoping review of 24 studies in Human Factors found promising results particularly for AR-based training, while VR-based training was mostly, but not consistently, as good as traditional training. A controlled comparison in Applied Ergonomics (2023) found no difference between AR, VR and video training in completion time or error count, but VR training came with a higher perceived workload and lower usability ratings. For a plant, that points to AR work instructions at the real station, and to VR only when the line, the machine or the hazard is not available to train on.

AR and VR for remote collaboration

Remote collaboration splits into two jobs that suit different tools: helping someone who stands at a real object, and meeting around something that exists only as a model.

Remote assistance with AR. A technician points a phone at a machine, and an expert elsewhere places arrows and drawings that stay attached to the real parts. Microsoft Teams supports this as spatial annotations: a desktop user can place 3D arrows or draw in a mobile user's space, on iPhones and iPads with ARKit or Android devices with ARCore, provided at least one person on the call holds a Dynamics 365 Field Service, Guides or Remote Assist licence. No headset is needed: the technician's own phone is the AR device, so there is no hardware to ship before the first call.

Shared design reviews and meetings in VR or MR. For a 3D design, a shared session keeps one copy of the model and each headset draws it from its wearer's position, so people can point at the same bolt. With visionOS 26, announced in June 2025, Vision Pro users in the same room can share an experience and add remote participants through FaceTime; Apple cites Dassault Systèmes' 3DLive app for reviewing 3D designs with remote colleagues.

For ordinary meetings, the evidence is sobering. In a study published on arXiv in 2022, 16 people worked in VR for a full week, eight hours a day, and rated it significantly worse than their normal desk on most measures, with concerning levels of simulator sickness; two dropped out on the first day because of migraine, nausea and anxiety. It is one small study, and the authors saw signs that people gradually got past their first discomfort, but it is a reason to keep the video call as the default and use headsets for the sessions that need 3D.

Which workplace XR products still exist in 2026

Several enterprise products ended in the last two years, so check the lifecycle page before you build a process on one:

If Remote Assist or Guides runs your field service today, plan the move before December 31, 2026 and export your guides from the Guides PC app, as Microsoft advises.

What AR and VR cost

Hardware is the visible cost, not the only one. Headset prices in the table above run from $349.99 (Quest 3S) to $3,499 (Vision Pro), and Vision Pro users who wear glasses need ZEISS optical inserts ($99 for readers, $149 for prescriptions), because glasses do not fit inside it. A realistic budget also covers:

1. Content. A simulation of your own machine or procedure is a software project in its own right; PwC's commercial study put its VR course at up to 48% more investment than a similar classroom or e-learning course, so the economics depend on how many people you will train.
2. Device management. Headsets need enrolment, updates, app deployment and a way to wipe lost units. Meta made HMS licences free in February 2026, and visionOS 26 added team device sharing, which stores each person's eye and hand data, prescription and accessibility settings on their iPhone so a pool of headsets can be shared.
3. Time and spares. Fitting, charging (rated battery life is 2.2 to 2.5 hours on Quest 3, Quest 3S and Vision Pro), facial interfaces, and supervisors who run sessions.
4. The systems around the headset. Sign-in, a content library, completion records and reporting to your learning platform.

That last item is easy to underestimate. Besides the headset app, a training programme needs a portal staff sign in to, internal tools that track who trained on what, and APIs that connect them to HR and learning systems. That is ordinary custom software development: portals and internal tools, with the APIs and back ends behind them.

Comfort, safety and accessibility limits

Cybersickness is the first limit to plan for. A 2020 meta-analysis of 55 studies (3,016 participants) found the pooled Simulator Sickness Questionnaire score across all content was relatively high compared with recommended cut-offs, with gaming content scoring highest and locomotion, visual stimulation and exposure time all influencing symptoms. Design for it: short sessions, seated or standing content without artificial movement, and an easy way to stop.

The makers set their own limits:

• Age. Apple says Vision Pro should not be used by children under 13. Meta requires a minimum account age of 10 for Quest (it can vary by country), and its Quest 3S warnings ask people who are pregnant, elderly or have binocular vision abnormalities, psychiatric disorders or heart conditions to consult a doctor first.
• Seizures. Meta's warnings state that about 1 in 4,000 people may experience severe dizziness, seizures or blackouts triggered by light flashes or patterns.
• Symptoms and breaks. Apple tells users to take regular breaks and stop at once on nausea, dizziness, headache, eye strain or blurred or double vision, and to feel reoriented before doing anything that needs balance.
• High-consequence work. Apple states that Vision Pro is not intended for use where a device failure could cause death, injury or severe environmental damage, so it should never be the only safeguard on a hazardous task.

Accessibility needs deliberate work. Vision Pro includes VoiceOver, Zoom, Voice Control, Switch Control, Dwell Control and Pointer Control, and the W3C's XR Accessibility User Requirements (a Working Group Note from August 2021) lists what XR content itself must offer: interactions that do not require particular body movements, voice commands, captions you can customize, a mono audio option, and avoiding triggers for motion sickness and seizures.

AR or VR: which one fits

Start from the job, not the device:

Then run a pilot that can fail cheaply:

1. Pick one task with a measurable outcome: errors, time to competence, repeat visits or travel days.
2. Measure the current method first, with the same people and the same measure.
3. Test the simplest device that works: a phone before a headset, off-the-shelf content before a custom build.
4. Track sickness and dropouts as seriously as the learning outcome, and keep a non-headset path.
5. Check the vendor's lifecycle (support dates, licence changes) before you build on it.

Some AR needs never require an installed app. Our augmented reality programming guide explains how AR runs in the browser with WebXR and how a 3D model on a product page can open in a phone's AR viewer, and how native AR apps are built when you need more. A 3D viewer on a web page still has to load fast on a phone like the rest of the page, which is how our web design and development team builds sites and web applications: held to Google's Core Web Vitals thresholds on real phones.