Ph.D. Candidate Andrew B. Park Is Building an AI Tool to Measure Ladder Safety, Not Guess at It

Falls are the leading cause of death on construction sites, accounting for more than a third of all fatalities, and ladders are the leading source of those falls.

That statistic is what drives research from Andrew B. Park, an NYU Tandon Ph.D. candidate in Assistant Professor Mohamad Awada’s lab. Park is building an AI tool that can look at a photo of a ladder and determine whether it's actually safe.

The problem, Park found, is that today's AI models aren't equipped for the job. Safety rules from the Occupational Safety and Health Administration (OSHA) and the American National Standards Institute (ANSI) come down to precise measurements, like the angle a ladder leans at, or how far its base sits from the wall.

While one recent study found that a leading AI model correctly caught 89% of real ladder violations, that seemingly impressive result falls apart on inspection. The model actually flagged so many photos as unsafe that only 18% of its violation calls held up. It wasn't judging ladders. It was guessing "unsafe" almost every time.

"No other study has trained AI to this level of precision on construction sites, especially for ladders," Park said. "That's the contribution: the system doesn't just look at a ladder. It measures it."

His solution, a project he calls Safety-LADDR, is designed to replace that guesswork with an actual measurement. Working with Awada, Park is training a system that takes a single site photo and checks it against the specific safety standard it must meet, the same way an inspector would, citation and all.

"He guides the way I should think, not just what I do," Park said of Awada.

Getting there hasn't been simple. No library of real construction photos exists with the precise measurement labels the tool needs to learn from, and collecting that data by hand, on real job sites, would be slow, expensive, and dangerous.

So Park has spent much of the past year building the training data himself: photorealistic, computer-generated construction scenes in which every angle and distance is known exactly. It's painstaking work, but it's what makes the eventual tool trustworthy rather than just plausible-sounding.

The stakes go beyond one project. Park's broader research, including a survey of construction-industry professionals that he led, found that the industry wants AI-assisted safety tools, but doesn't trust ones that merely describe what they see. It wants tools that check real numbers against real rules, which is exactly the gap Safety-LADDR is meant to close.

Park came to this work by way of architecture, not computer science. He spent years designing buildings before shifting into AI research, a background he says gives him a feel for how construction sites actually operate, not just how to model them.

Wherever the next step takes him, Park is certain about one thing: he wants the work to leave the lab.

"I want this in the hands of the people who carry the cost of safety: construction industries, municipalities, insurers, agencies, contractors," he said. "I'll keep pushing. There are more tools to build beyond ladders."

For this research, Park has been named a recipient of the NYU Urban Doctoral Fellowship through the NYU Urban Initiative, which recognizes doctoral students across NYU whose research advances understanding of urban challenges.

"Only a small number of students across the whole school are chosen each year," he said. "I was genuinely honored."