Skip to Main Content

Despite the design and engineering sophistication required to build a tall building, the construction industry still struggles to get projects done on time. That’s because humans, not computers, do most of the manual labor, and people are prone to inefficiency.

Artificial intelligence can help.

Under the guidance of UIUC Prof. Mani Golparvar-Fard, a talented team of five undergraduate students from three DPI partners created a complex data set of building materials, such as wood and tile. The data set will be used in a startup’s AI model so that its software automatically identifies delays and progress as buildings go up.

Construction relies on material deliveries, manual labor by lots of people and coordination between trades, which almost inevitably results in expensive holdups, scheduling snafus and work errors.

“The biggest part of our project was to automate some of the progress-tracking — making sure that whatever is happening on a construction site is actually on schedule, and if it’s not on schedule, we wanted to track how behind it was,” said Devyani Gauri, an artificial intelligence major at the Illinois Institute of Technology.

The students were part of the first group of DPI Research Scholars, a program placing junior and senior engineering, computer science, data science, or data analytics students from the University of Illinois Chicago, University of Illinois Urbana-Champaign, and Illinois Tech into small groups to work on specific tech projects identified by DPI science teams.

The Research Scholars program was put together on the fly during COVID as a substitute for more traditional internships that had to shut down, which potentially robbed students of valuable real-world experiences.

“We got creative,” said DPI corporate engagement lead Jeffrey Donne.

Administrators looked at DPI’s science teams — research initiatives run by professors — and decided to match up student workers with the teams to solve problems using artificial intelligence. The students also took a business course focused on entrepreneurship to give them a broader perspective and further encourage these promising future professionals to consider tech careers in Illinois, a primary DPI goal.

“It gave the students a very unique opportunity,” Donne said. “Undergraduates don’t often get to do research. I think they surprised some of the professors by the quality of the work they were able to accomplish.”

There are nine DPI science teams. Six of them assigned projects to groups of Research Scholars, including ones focused on autonomous farming, autonomous vehicles, and brain research.

Then there was AI for Construction, whose project lead is Golparvar, associate professor in the Department of Civil and Environmental Engineering at UIUC. Golparvar is also co-founder of Reconstruct Inc., a startup that uses AI to help construction managers monitor progress and make better decisions.

The assignment to students was to integrate AI into a robotic 3-D mapping system for construction. The system would use machine learning to analyze construction progress and problems in order to improve the speed of assessing a project’s schedule status.

The AI system’s job is to compare construction plans with the chaotic reality of a building site and determine, using collected video imaging whether, for example, the plumbing has been installed, if the beams in the sub-basement look good, and, hey, someone forgot to put electrical boxes on the 42nd floor.

The Research Scholars were given several specific tasks, key among them creating a data set of construction materials — concrete, bricks, tile, etc. — to prepare the AI system for analyzing sites. That work required extraordinary computational and organizational effort, such as giving the computer 151 different lighting conditions to simulate how the materials appear in real-world construction environments.

“We trained a model, and then when we feed in the actual real-world data, we will be able to recognize which object has been worked on,” Yan Sun, another of the research scholars, explained in a presentation. The other three team members are Clark Chen, Yuanrui Chen and Pratik Patel.

“It was astounding work,” said Juan Diego Núñez Morales, mentor to the team and research assistant to Golparvar. “What they were solving was a critical part of rolling out AI solutions: The collection of massive amounts of high-quality annotated data needed to train the model.”

They did this project amid COVID, which meant they collaborated entirely via remote meetings, relying on a Google Doc to keep track of their progress. Gauri, an international student from India who will continue studying AI in graduate school at Northwestern, said the project opened her eyes to a new field of research.

Perhaps because construction work is messy, it doesn’t have the same high-tech allure as web development or brain research. Now Gauri and her fellow students know more.

“It did give me a whole new perspective on what AI can do, even in fields you might not think of as AI-predominant.”

By Michael Lev