The most dangerous worker in a highway work zone might be the one who’s been there longest.
That’s the premise behind research at Texas A&M by Namgyun Kim, an assistant professor in the Department of Construction Science, working with Brian Anderson at Texas A&M and Yongcheol Lee at Louisiana State. The team is using VR simulations, brain-activity monitoring, field observation and AI-driven augmented reality to study a phenomenon they call habituation.
Habituation is a perception problem, not an attitude problem
Workers who see the same warning signs, the same alarms and the same passing traffic every day stop registering them. It isn’t carelessness. It’s how attention works. And it has been nearly impossible to measure, which is why safety programs have historically treated it as a briefing problem.
Wearables and AI change that. Kim’s group can now watch, objectively, when a worker’s attention stops landing on a hazard, and deliver targeted training before that pattern turns into an incident.
The scale of the problem justifies the effort. NIOSH data shows struck-by incidents caused roughly 1,800 fatalities and more than 167,000 nonfatal injuries between 2011 and 2021.
Generic VR isn’t enough
Kim is blunt about the limits of what’s on the market. “Most VR programs are built around a fixed set of scenarios,” he told Construction Dive, and “a generic VR scenario often cannot capture the unique hazards of a specific project or task.”
That’s the gap generative AI fills. It “can dramatically reduce the time and cost required to create new training scenarios,” Kim said, which is what makes project-specific simulation economically plausible for the first time.
His headline claim is a challenge to the industry: “VR- and AI-based safety training should not be viewed as an experimental technology anymore.”
The ROI problem nobody has solved
Here’s the honest part, and Kim says it himself: “Safety incidents are relatively rare events, making it difficult to prove that any single training program directly prevented an accident.”
That’s the wall every safety technology hits. You can’t run a controlled trial on fatalities. Contractors buying this are buying a plausible mechanism, not a demonstrated reduction, and the research doesn’t yet publish a percentage improvement.
Which is a reason to be careful with vendor claims, not a reason to ignore the work. Habituation is real, it’s now measurable, and the alternative approach — another toolbox talk about looking both ways — has had a hundred years to work.
For a sense of where this technology meets live highway work, see our listing for the Prairie Creek plant expansion, a job that runs entirely inside an operating facility. NIOSH’s struck-by data is the baseline any claim should be measured against.