As potentially lucrative as the construction industry is, it is fraught with challenges. No two sites and projects are alike. Labor forces almost assuredly will vary from job to job. Heavy equipment is constantly in motion, posing potential dangers that must be recognized before accidents happen.
 
The advent of unmanned aircraft systems in recent years has helped construction site managers stay ahead of potential pitfalls.
 
However, UAS flights at times can raise more issues than they resolve. Camera and information payloads are sophisticated and affordable to the point where they actually could inundate operators with too much information. Gleaning the vitally important data quickly can be nearly impossible for construction site managers, who already are beset with enough to do.
 
Enter artificial intelligence. Autodesk Inc., a software company based in Rafael, California, is working with client construction companies to develop AI systems that cull out the necessary information from the mountain of data produced by robotic devices.
 
“Humans can do a whole lot of things really well. But what we can’t do is storage, computation, and data analysis,” says Tristan Randall, strategic projects executive at Autodesk. “A computer tends to do a better job of identifying patterns and data, correlating different data sets, and doing things like searches across different domains.”
 
Building upon its reputation for providing industry-standard software to the construction companies worldwide since the late 1980s, Autodesk has spent much of the last few years developing products aimed at equipping customers with the tools they need to manage the piles of data they receive.
 
Cross pollinating
 
To do this, Autodesk adapts the more than 200 products developed for its business lines — infrastructure, manufacturing, media and entertainment — to the construction industry’s needs.
 
“We can cross pollinate,” Randall says.
 
For example, he says, software that is taking off in the technology sector — which tends to allocate a good portion of resources toward research and development — could prove useful for construction companies, which do not. Likewise, architects and engineers who want to generate high-quality renderings at the click of a button can use visualization capabilities from the company’s media and entertainment division.
 
Autodesk has since formed an internal panel, tasked with seeking future projects in which AI could enhance the process of finding solutions to their construction customers’ problems.
 
“We can build algorithms that can perform discrete tasks on data,” Randall says. “The ultimate goal of using AI is to help humans gain insight to the data they have, so that they can do what humans do, [only] better.”
 
Managing the information provided by drone flights provided a good place to start.
 
“Drones now are at the point where they’re cheap enough, with onboard systems like obstacle avoidance, GPS and tracking capabilities good enough so those basic aspects of technology have essentially been democratized. Any project can leverage drones for value,” Randall says.
 
Too much value, perhaps. Randall says a drone on a mapping mission can generate anywhere from 100 to 1,000 individual images, taking up anywhere from 10 to 20 megabytes in storage. Producing an ortho-mosaic map, one that “stitches” together a multitude of images, could generate files ranging from 500 megabytes to three or four gigabytes. Three-dimensional images easily can take up more than a gigabyte.
 
In total, a construction-site foreman may be presented with something in the neighborhood of 20 gigabytes of data in a single week. Randall has talked to some construction site customers whose on-site servers may be limited to 20 gigabytes a month.
 
Nevertheless, Randall says, the useful data is in there somewhere: How much dirt has been moved since a month ago. How many active pieces of equipment are on a site at a given time. Which subcontractors are on a site, and what tasks have they completed.
 
A construction project, by definition, is never the same at any one point. Equipment, cranes, people and supplies are constantly coming in and out. Waste is getting moved off site. Studies have shown that somewhere between 15 and 20 percent of the materials on any given construction site are wasted, making construction one of the biggest contributors to solid waste in both the U.S. and the United Kingdom, Randall says.
 
A contractor’s ability to manage such activity plays a heavy role in determining whether or not the enterprise is profitable.
 
“The solution is not to fly less or capture less data. It’s to use AI to extract the information we need, so that we’re not trying to match huge data files,” Randall says. “We’re pulling information from them that we need, using these algorithms.”
 
The UAS cameras have the answers. Artificial intelligence helps deliver those answers faster, Randall says.
 
“The stakes are high. Construction is an industry with a low margin,” Randall says.
 
Barriers to innovation
 
Because of competitive pressure, the profession is beset with what Randall calls stagnating labor productivity. Construction teams rarely work together on the same project twice. Likewise, no two projects are ever alike. Additionally, the construction business invests relatively little in technology when compared to other commercial enterprises.
 
“It’s not that the construction industry doesn’t want to innovate,” Randall says, “but they face barriers to innovation unparalleled to other industries.”
 
Still, construction is a massive industry, generating somewhere between eight and 10 trillion dollars a year worldwide. In the U.S. alone, more than $1 trillion is spent on construction annually.
 
“The opportunity is significant. That’s why we’re so focused on how AI technology can be applied,” Randall says.
 
The company is well into a project begun several years ago, which it calls the BIM360 Forge platform — essentially, cloud-based technologies that it created specifically for the construction industry. (BIM stands for building information modeling.) Using Forge, third-party companies can incorporate their own technologies on top of it.
 
“This is important for us, because Autodesk doesn’t build hardware. We’re a software company,” Randall says. “We need hardware partners we can work with, to ensure that we can address the full work flow from site to data and be able to offer that to our customers.”
 
As of now, two hardware companies — 3D Robotics of Berkeley, California, and San Francisco-based DroneDeploy — have joined with Autodesk, Randall says, with more expected to follow suit. The collaboration is making it easier for the companies to extract rooftops for insurance claims, count cars in parking lots, see through vegetation, and figure out how much material is sitting in stockpiles on sites.
 
Additionally, Autodesk is developing a series of desktop tools, with practical applications in civil engineering drafting and documentation, among other potential tasks. With the three-dimensional design tool the company calls Autocad, users at desktop stations can automatically sort out vegetation, signs, or any vertical structures from drone-produced images.
 
“We can use machine learning [AI] to train a system to more effectively extract those ground surfaces. Companies can actually use a drone to capture a site, and then extract a terrain model for use as base maps or contour maps — automatically,” Randall says.
 
Traditionally, doing so would require having someone go into files and manually pull out features they do not want included in a final product.
 
“The advantage of machine learning is that you can train it with whatever data sets you like. We’re exploring how we can apply it more broadly than what’s available today,” Randall says.
 
Autodesk also is developing systems that would be able to extract data directly from images coming from any number of existing platforms that provide them, such as Google’s TensorFlow or the one inherent to IBM’s Watson. The company also is considering building its own image-based AI product.

Below: A look inside Autodesk's San Francisco office. Photo: Autodesk