Can AI Learn the Physical Wisdom of an Expert Mason?

Masonry and construction are demanding fields, requiring constant and repetitive motions that can lead to fatigue, wear and tear, and musculoskeletal injury. Those injuries represent billions of dollars in healthcare costs each year, and the physical toll can lead to young builders leaving the profession or experts retiring early.

A team of researchers at the University of Waterloo in Ontario, Canada, believe that artificial intelligence (AI) can help mitigate those risks—not by replacing humans with machines, as is often feared, but instead by training those humans to work smarter and safer.

“The people in skilled trades learn or acquire a kind of physical wisdom that they can’t even articulate,” said Carl Haas, a civil and environmental engineering professor at the university. “They’re basically doing the work twice as fast with half the effort, and they’re doing it with higher quality.”

Because those workers cannot explain the work habits that make them both durable and efficient, Haas and his colleagues recruited 21 masons with varying experience levels and studied their motions. Outfitted with Xsens MVN motion sensor suits, the masons built a wall with concrete blocks while the research team assessed how much stress they put on their bodies, particularly their joints. The data confirmed that the more experienced workers accrued the least physical stress while building the wall fastest.

In a second study, the sensors were used to capture the movements of the masons, which were used to develop a support vector machine (SVM) supervised learning algorithm to find patterns of motion. Using the stances captured by the sensor suits and sorted by the algorithm, the researchers put together “pose books” of effective motions.

About 70% of the motion sensor data collected was used to train the algorithm to sort between “expert” and “inexpert” work poses: The remaining 30% of data was used to test it. The algorithm was over 90% accurate in 2 different classification scenarios (binary and inter-group multiclass).

“These pose books can be used to develop systems to train novice workers to work in ergonomically safer, more productive ways,” the study says. Experts often don’t follow “standard ergonomic rules taught to novices,” as a statement from the university puts it: They develop their own means of motion over time, for example, by doing more swinging than lifting.

“Skilled masons work in ways we can show are safer, but we don’t quite understand yet how they manage to do that…Now we need to understand the dynamics,” said Haas, who led the work alongside systems engineering professor Eihab Abdel-Rahman.

The work is ongoing, and the team is looking to develop a training program that uses video recordings and motion sensor suits to give apprentices instantaneous feedback on their motions.

The study, “Identifying poses of safe and productive masons using machine learning,” was published recently in Automation in Construction.