Giant robots made by Austin startup Icon began 3D-printing an entire neighbourhood of suburban homes in late 2022; on Tuesday, the company introduced an AI architect to design them. It’s a technological leap that Icon says could one day spell the demise of cookie-cutter suburbia, ushering in an era of lower cost, lower carbon and climate-resilient homes in wildly novel forms.
That’s the promise, at least, of a suite of new technologies Icon previewed for Bloomberg Green as its machines finish construction on the world’s largest 3D-printed neighbourhood, a 100-house Austin subdivision at Wolf Ranch for home giant Lennar Corp. The innovations, formally introduced Tuesday at SXSW, range from the AI architect to low-carbon concrete and a new robot that can 3D-print two-story buildings, roofs and foundations included.
ADVERTISEMENT
CONTINUE READING BELOW
“There’s no reason to build ugly-ass spec homes anymore,” Icon founder and chief executive officer Jason Ballard tells me at the company’s headquarters near downtown Austin (while wearing his customary white cowboy hat). “We don’t just need to make housing more affordable; it actually also needs to get better.”
Icon has spent 18 months developing its artificial intelligence programme — named Vitruvius after an ancient Roman architect — which uses a chatbot to converse with would-be homeowners about their dream projects. After the user responds to a series of prompts, Vitruvius offers several versions of a house, complete with exterior and interior renderings and floor plans.
“It has an intuitive awareness of both the laws of physics and a bit of building code and building possibility,” Ballard says. When he asked Vitruvius to design a 3D-printed treehouse, for instance, it included a support column as it knew a tree alone couldn’t bear the weight.
For now, humans are still needed to turn Vitruvius’s plans into buildable reality. By year’s end, the program will be able to produce its own construction schematics, according to Ballard. He says Vitruvius is at least a year away from being able to draw up full construction documents, permit applications, budgets, bills of materials and a building schedule — an ambitious timeline for an untested technology tacking a complex interplay of tasks. But if those processes can be automated, Icon estimates the software could shave at least $100 000 from the price of a home.
Getting there, though, will necessitate acquiring untold quantities of data and training the program on a plethora of factors, including building codes, the geography of construction sites and perhaps even the peculiarities of homeowner association rules. Icon is counting on thousands of Vitruvius users to help refine the system (free for now) once it launches. Judging by the requests of about 100 beta testers, the AI will have plenty to work with: Design asks so far include everything from Barbie dream homes to Hobbit hideaways.
“Let’s play. What do you want?” Ballard asks me as he launches Vitruvius on a conference room screen and a chat window pops up. “I’m here to help you design a home,” it says. It’s something few prospective homeowners ever hear. Unless you have the budget to hire an architect, options dwindle to mass-produced models offered by suburban developers or what’s been built before.
I ask Vitruvius to show me a 1 500-square-foot, three-bedroom, two-bath home. I specify a contemporary style with expanses of glass. Location: the beach town of Bolinas in Marin County, California. “Sounds like you’re envisioning a contemporary one-story home with an open feel,” Vitruvius responds. “Before we proceed with the design, can you tell me more about your lifestyle and how you envision using the space in your home?”
I tell the AI that two people and a dog would live in the house and that I want a seamless flow between home and nature, plus a work space. “Considering your location in Bolinas and desire to integrate with nature, would you like the design to include specific outdoor features such as a deck patio or garden areas?” it asks.
I would. Within seconds, Vitruvius displays a collection of stunning homes and floor plans in a setting that resembles coastal California. The homes include conventional construction as well as a 3D-printed model that takes advantage of the technology’s ability to create undulating walls that blend into the landscape. (Vitruvius always includes at least one 3D-printed version of a home, which Ballard says would-be homeowners seem to favour.)
Developers typically tamp down generative AI programs’ tendency to “hallucinate” or provide information that’s untethered from reality. Icon initially did the same until, “We realized when you turn hallucination to zero, you get the most offensive version of cookie-cutter developments,” Ballard says. “You actually need a little bit of hallucination to get 3D-printed tree houses.” The key is balancing creativity against what can practically be built. “You can’t hallucinate construction budgets,” he quips.
Ballard credits the inspiration for Icon’s AI architect to a chance encounter with OpenAI CEO Sam Altman at a conference two years ago, before the release of ChatGPT triggered an AI gold rush. Ballard told Altman about his idea for architecture software that might eventually add AI.
“He said, ‘Your timeline is wrong. You should start building it as an AI system right now,’ and gave me some tips and pointers,” says Ballard, who sent a beta of Vitruvius to Altman. A representative for Altman did not respond to a request for comment.
Maria Paz Gutierrez, an associate professor of architecture at the University of California at Berkeley, says 3D printing “in principle can offer unique advantages to address affordable housing prices issues because it can substantially decrease building time and potentially material usage.”
“In a general sense, AI is bound to change the way in which not only we design, but how we build, and ultimately how we live our day-to-day lives in spaces,” she says.
So is architect the next human job under existential threat from artificial intelligence? Ballard says Vitruvius will relieve architects of onerous design duties like drafting construction documents. Icon is also offering architects an opportunity to profit from its 3D-printing tech if they submit plans to the company’s new catalog of ready-to-print home designs, CodeX.
ADVERTISEMENT
CONTINUE READING BELOW
“Every time we build one of your designs, we will send you a check so that people have access to good architecture again without paying $30 000, $60 000, $100 000 or $200 000,” Ballard says. Icon’s CodeX program will pay architects 1% of the construction costs for any home built using their blueprints and will help them design for 3D printing.
CodeX is launching with 63 plans for homes, ranging from under $99 000 to well over $1 million. The catalogue includes some 40 designs from famed architects Bjarke Ingels Group that take advantage of 3D-printing to create avant-gard wildfire- and storm-resilient homes. (One model, for instance, mounts kitchen cabinets high off the floor to minimize flood damage.)
But Icon’s ability to actually execute on Vitruvius and CodeX designs depends on the company’s next-generation robot, called Phoenix. During my visit, a prototype works across the street to build a large Gaudí-esque structure whose walls ripple across the lot like an infinite wave.
The robot behind Icon’s 3D-printed neighbourhood, Vulcan, consists of a crossbar that moves up and down between two towers that travel along fixed rails straddling a building site. A nozzle attached to the crossbar extrudes a proprietary concrete mixture called Lavacrete, which the robot lays down layer upon layer to form the exterior and interior walls of a one-story building.
The nozzle on Phoenix, on the other hand, sits at the end of a 23-foot-long arm attached to a small platform with tank-like treads that can freely rumble around a building site. In Austin, Phoenix’s arm swoops down to the curving 100-foot-long structure, adding layers of Lavacrete to the second story of what will be a 27-foot-tall building when completed.
Phoenix’s height and manoeuvrability gives it the capacity to build multi-story structures in more free-flowing designs and further automates construction. Where it took at least a day for workers to set up Vulcan’s rail system at each individual home site in Wolf Ranch, Ballard says Phoenix is ready to print upon arrival and can quickly move from job to job.
In another improvement, Phoenix can build roofs and foundations. Vaulted or dome roofs are printed on top of a structure, while panels for flat roofs are printed on the ground and installed later. To build a foundation, Phoenix prints the forms that encase the poured concrete. While workers had to manually install rebar to reinforce the walls printed by Vulcan, a spool of galvanized steel cable is extruded inside each layer of concrete Phoenix applies.
Icon estimates that Phoenix will cut the build time for a 3D-printed home in half, to one week, while reducing the cost of the wall system by 30% compared to conventional construction. “Over time, the cost of building with a robot should approach $3 an hour while the cost of human labour is going to be approaching $20 or $30 an hour,” Ballard says.
The climate cost of Icon’s homes is also falling. While its 3D-printed homes generate little construction waste, concrete’s carbon footprint makes the homes’ emissions substantially higher than wood-frame alternatives. On Tuesday, the company unveiled a low-carbon concrete formula called CarbonX that brings the emissions of a 3D-printed home 2% to 6% below those of a same-size wood-frame building over their lifespan, according to an analysis conducted by Icon and the Massachusetts Institute of Technology.
“This 3D printing allows us to improve the productivity of concrete construction, and newly engineered materials and concrete mix design like Icon’s reduces emissions associated with the production of concrete,” says Hessam Azarijafari, deputy director of MIT’s Concrete Sustainability Hub.
A world in which Icon’s home-building robots execute plans shared directly by its AI architects remains a ways off, but Ballard says that’s the likely future. “Humans will be in the loop at multiple points, but directionally I think that’s correct,” he says. “An AI system isn’t limited in its capacity for memory and intelligence and creativity.”
© 2024 Bloomberg