Despite the enormous promise of robotics to change life as we know it, today’s robots are largely dumb and inflexible as they confront a complex and ever-changing world. “Nobody was doing it right, and it would be huge if we could do a better job,” says cognitive scientist Gary Marcus.
So Marcus, a former NYU professor and founder of another startup called Geometric Intelligence that was bought by Uber, teamed up with roboticist Rodney Brooks, cofounder of iRobot and co-inventor of the Roomba, to start a new company. Their Palo Alto, California-based startup, Robust.AI hopes to bridge the gap between what robots can do today and their promise by building what they call “the world’s first industrial-grade cognitive engine for robotics.” Its goal is to help companies build and rapidly deploy semantically aware robots that could ultimately work on construction sites, care for elderly, make fully autonomous driving a reality, or do other complex tasks.
“We want to build a cognitive engine that will allow people to set up robots faster and have more faith in their reliability,” Marcus says. “Instead of hiring 40 Ph.D.s and taking three years, they should be able to take six months with maybe five Ph.D.s. That’s the vision of what we are trying to build—that’s a long-term vision.”
Operating largely under the radar since its 2019 founding, Robust.AI said today that it had raised $15 million led by Jazz Venture Partners and joined by previous investor Playground Global, bringing its total funding to $22.5 million. Marcus, the company’s CEO, declined to disclose the company’s valuation other than to say that “it’s not obscene, but it’s healthy.” Other investors in the company include angel investor Esther Dyson, Veritas software cofounder Mark Leslie, Skype cofounder Jaan Tallinn and Silicon Valley entrepreneur Steve Blank.
With the new funding, Robust.AI, which has some 25 employees, intends to build out the software, focused first on social perception and navigation, and to bring on salespeople to begin to commercialize its offerings. Marcus says that the company is currently doing a pilot project with a customer he declines to name, and is “in conversation with several other partners.” As Robust.AI is a software company, its partners will need to provide the hardware. Marcus hopes to begin shipping product in 2021. Building a company like this is a long and difficult process, and, even if all goes according to plan, it will take time before its software is broadly available under license.
“We want to build a cognitive engine that will allow people to set up robots faster and have more faith in their reliability,” says Robust.AI’s Gary Marcus.
The roots of the company go back to conversations that Marcus, 50, and Brooks, 65, began having about the problems with artificial intelligence and how it’s been applied. “We think there’s a lot more to AI than just big data sets,” Brooks says. “I’m of the opinion that it’s almost a fetish.”
The two connected after Marcus was working on his latest book, Rebooting AI, coauthored with NYU computer science professor Ernest Davis, about the limitations of artificial intelligence. He sent the chapter on robotics—which delved into the gap between the potential of robotics and its reality—to Brooks. “He said, ‘You’re way too nice,’ which is not something that people say to me,” Marcus recalls.
Research on the book had also sparked the idea for the company, and he floated the idea to Brooks of becoming an adviser. “It was sort of a flyer. I was in awe of what he’d achieved,” he says. “Then I spent three months recruiting him.”
At the time, Brooks was winding down collaborative robotics startup Rethink Robotics, which shuttered in October 2018. He soon signed on as a cofounder and chief technology officer. The company’s three other cofounders have similar stellar pedigrees: Mohamed Amer previously led AI and machine learning projects at SRI International; Anthony Jules had been chief technology officer of Formant.io, an intelligence platform for robot fleets and a senior product manager at GoogleX; and Henrik Christensen is a professor at UC San Diego and director of its Contextual Robotics Institute.
“We wanted to get a software platform that would help robots work in the real world, and not in these constrained environments,” Marcus says. “There is nothing off the shelf where if you want to build a domestic robot you can get started the way you can if you want to get a game engine and build a video game.”
With a game engine, he explains, video game developers can let the software work out the tedious problems, like where objects will collide, and focus on the creative aspects of the game. Robotics, he believed, should have the same foundation, rather than requiring engineers to hack together a mix of Robot Operating System, or ROS, and other programs tailored to specific problems.
As the use of artificial intelligence and automation spreads to industries that previously hadn’t relied on it, the need for such foundational software should increase. “A lot of companies that didn’t previously think about having AI in their systems will need it or want it,” Brooks says. “They can’t all have world-class research labs in AI. It has to be a commodity for them.”
In the short term, Marcus says, Robust.AI is focused on social navigation, that is the simple problem of how to go around people or objects and not interrupt or be obtrusive. Longer term, he has ideas for how to process language more accurately and efficiently so that, for example, a request to put objects in a closet does not result in the robot hacking those objects to bits to make them fit.
“We wrote this great chapter [in Rebooting AI] on why the approach to language that deep learning was taking would never work, and I stand by that claim,” he says. “The fundamental problem is that we don’t say everything we need.”