Industries / Industrial equipment and services / Industrial services / Safety training
Simulators and virtual reality
Simulators and virtual reality (VR) let workers rehearse dangerous jobs and emergencies in a copy of the workplace, from a replica control room to a headset that shows a computer-drawn site.
6
Milestones achieved, from 1960 to 2026.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1960
A reactor rehearsed by computer
At a computer conference in May 1960, Samuel N. Irwin and Robert R. Kley presented an analog computer that served both as a tool for analyzing the systems of the Enrico Fermi Atomic Power Plant and as a facility for training its operators. The IAEA dates the first computer-based full-scope simulators, copies of a whole control room driven by a model of the plant, to the 1970s, when limited computing power kept their models simple.
Credit
Samuel N. Irwin and Robert R. Kley
1985
Simulators matched to the plant
An IAEA report of 2021 says that operator training on a vendor's full-scope simulator that differed in significant ways from the real control room was one of the contributing factors to the 1979 accident at Three Mile Island Unit 2. In 1985 ANSI/ANS-3.5-1985 became the first comprehensive standard for the functions of simulators used to train and examine nuclear plant operators. By the 1990s, the IAEA says, control room staff in virtually all countries with nuclear plants trained on simulators, more than 90% of them on plant-referenced simulators built to match their own plant.
Lessona simulator that differs from the real controls can teach the wrong moves.
Credit
American Nuclear Society; nuclear regulators and plant operators
2013
VR helps on some tasks
Rafael Sacks and Ronen Barak of the Technion and Amotz Perlman of Bar-Ilan University in Israel trained 66 people in construction safety, half in a classroom with visual aids and half in front of an immersive VR "power wall", a room-sized 3D screen, and tested them before, just after and a month later. VR training did better for stone cladding and cast-in-place concrete work but not for general site safety, and it held trainees' attention better.
Credit
Rafael Sacks, Amotz Perlman and Ronen Barak
2019
VR no better than slides
Johannes Leder and colleagues at the University of Bamberg and Chemnitz University of Technology in Germany compared safety training in immersive VR with a PowerPoint presentation containing vivid film scenes, in two experiments with 53 apprentices and 68 participants. The first found effects on risk perception and risky choices but not on learning; the second, which also retested people after six months, could not repeat them. The authors concluded that the cost of VR did not seem justified for safety training.
Lessona cheaper method done well can match VR, so a test needs a strong comparison.
Credit
Johannes Leder, Tina Horlitz, Patrick Puschmann, Volker Wittstock and Astrid Schütz
2024
Gains found, rarely retested
D. Scorgie, Ruggiero Lovreglio and colleagues, mostly at Massey University in New Zealand, reviewed 52 studies of VR safety training published from 2013 to 2021, outside medicine and the military. Two meta-analyses, which pool the results of many studies, found that VR beat traditional training for knowledge gained and kept. Only 36% of the studies measured long-term retention, and 9.6% said which learning theory the training was built on.
Lessonmost tests of VR stop soon after training and measure what people know, not what they do.
Credit
D. Scorgie, R. Lovreglio and colleagues
2026
Meta ends business headset sales
Meta's program for business use of its headsets began as Oculus for Business in 2017, became Meta Quest for Business in late 2023 and was renamed Meta Horizon managed services in 2025; it offered commercial licenses, warranties and remote management of many headsets. From 20 February 2026 Meta stopped selling commercial versions of its Quest headsets and new subscriptions, cut existing subscriptions to $0 a month and set 4 January 2030 as the end of support, as it moved spending from VR to smart glasses, UploadVR reported.
Lessontraining built for one maker's headsets depends on that maker's plans.
Credit
Meta
Watch for
Extinguisher skills after VR
D. Scorgie, R. Lovreglio and colleagues taught 130 people a seven-step fire extinguisher procedure by immersive VR, by augmented reality (AR, computer images laid over the real room) or by video. The VR and AR groups did significantly better than the video group in a test with a real extinguisher, VR best of all, and kept more knowledge after three months. The real-extinguisher test came right after training; only knowledge was retested at three months.
Achieved whenanother team repeats the result, with the hands-on test also repeated months later.
Claimed by
D. Scorgie, R. Lovreglio and colleagues, Massey University, October 2026
VR for substation crews
J. M. G. del Pozo and E. R. Segovia of the Universidad Politécnica de Madrid report that at a Spanish infrastructure and energy company, 10 workers trained with manuals plus VR on a Meta Quest 2 headset averaged 0.813 on a first knowledge test against 0.707 for 10 trained with manuals alone, and also scored higher in a full-scale substation walkthrough and in fire drills. The groups were formed by work shift and subproject, not at random.
Achieved whena randomized trial with more workers, scored by assessors who do not know who had VR, finds a difference.
Claimed by
Del Pozo and Segovia, March 2026
Open
01
Hazard hunts in generated scenes
A test of whether trainees find the hazards in computer-generated site scenes as often as in real photographs. Scenes are generated from OSHA's severe injury narratives, each with a known list of planted hazards, and paired with real photos of similar work; trainees mark every hazard they see.
The briefCredit · unclaimed02
VR against practice, months later
A registered trial that teaches one procedure, isolating and locking out a machine on a training board with no power, by VR, by hands-on practice on the board or by video, then has assessors who do not know each person's group score them on the real board after one week and after three months.
The briefCredit · unclaimed
Open 01 · the brief
Hazard hunts in generated scenes
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A test of whether trainees find the hazards in computer-generated site scenes as often as in real photographs. Scenes are generated from OSHA's severe injury narratives, each with a known list of planted hazards, and paired with real photos of similar work; trainees mark every hazard they see.
Why it is still open
Trevor Neece, Mason Smetana and Lev Khazanovich turned 75 OSHA severe injury records into 750 images in May 2026; experts rated 81.1% of the single images acceptable for teaching, but no trainees were tested on them. A 2026 review of 65 studies of generative AI in construction safety found the evidence for field use and injury prevention limited.
Built with what exists
OSHA's Severe Injury Reports data, which hold a description of each amputation, hospital stay or eye loss reported since 1 January 2015; an open image model such as Black Forest Labs' FLUX.1 or Stable Diffusion; Unity to turn scenes into walk-throughs; later, Google DeepMind's Genie 4 world model, which draws a walkable scene from typed text but is so far open only to trusted testers (see the Note "A world that follows"); a web form or PsychoPy to run the test.
First step
Generate 20 scenes from fall and struck-by narratives in the OSHA file and have two safety professionals list the hazards each one shows.
Done looks like
100 generated and 100 real scenes, each hazard list checked by two safety professionals; at least 60 trainees; the difference in the share of hazards found between the two sets, with a 95% confidence interval; scenes and scores released.
Open 02 · the brief
VR against practice, months later
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A registered trial that teaches one procedure, isolating and locking out a machine on a training board with no power, by VR, by hands-on practice on the board or by video, then has assessors who do not know each person's group score them on the real board after one week and after three months.
Why it is still open
Only 36% of the 52 VR safety studies reviewed in 2024 measured long-term retention. The 2026 fire extinguisher trial tested real-world performance only right after training, and a Colombian trial of VR before work-at-height practice at SENA, the national training service, was published in 2026 as a plan without results.
Built with what exists
Consumer Meta Quest 3 headsets, since Meta stopped selling business versions in February 2026; Unity with OpenXR, the cross-maker headset standard, so the course also runs on other headsets; commercial lockout training boards; OSF to register the plan.
First step
Build a VR copy of one lockout training board in Unity and try all three methods on 15 volunteers.
Done looks like
At least 150 participants assigned at random to the three groups, at least 70% retested at three months, and step-by-step scores per group published with the registered analysis.